WEBVTT

00:00:00.000 --> 00:00:01.360
<v Bertrand>This is The Amp Hour Podcast.

00:00:03.000 --> 00:00:05.820
<v Bertrand>Recorded July 14th, 2015.

00:00:07.400 --> 00:00:09.220
<v Bertrand>Episode 258.

00:00:11.100 --> 00:00:14.340
<v Bertrand>An interview with Bertrand and Gerald of Audine.

00:00:30.000 --> 00:00:34.460
<v Dave Jones>Welcome to the Amp Hour. I'm Dave Jones from the EEV blog.

00:00:34.740 --> 00:00:36.780
<v Dave Jones>And I'm Chris Gammell of Contextual Electronics.

00:00:37.420 --> 00:00:39.220
<v Gerald Friedland>I'm Gerald Friedland of Audine.

00:00:39.520 --> 00:00:41.500
<v Gerald Friedland>And I'm Bertrand Irisu from Audine.

00:00:41.920 --> 00:00:43.700
<v Dave Jones>Hey guys, thanks for joining us.

00:00:43.980 --> 00:00:44.440
<v Chris Gammell>Thank you.

00:00:44.880 --> 00:00:45.280
<v Dave Jones>Thank you.

00:00:45.280 --> 00:00:49.240
<v Dave Jones>It's an interesting topic we've got today, voice recognition.

00:00:49.460 --> 00:00:50.140
<v Bertrand>I'm pumped about this.

00:00:50.880 --> 00:00:57.800
<v Dave Jones>Yeah, speaker independent voice recognition, which is even harder than regular voice recognition, is it not?

00:00:57.800 --> 00:01:01.140
<v Gerald Friedland>Yeah, that's a good question.

00:01:01.620 --> 00:01:06.400
<v Gerald Friedland>So speaker independent voice recognition these days is kind of the state of the art.

00:01:07.760 --> 00:01:10.280
<v Gerald Friedland>So there are different ways of doing it.

00:01:10.900 --> 00:01:16.860
<v Gerald Friedland>One way is to basically have acoustic models pre-trained with many, many speakers.

00:01:17.280 --> 00:01:21.480
<v Gerald Friedland>So that if you sound similar to one of them, it just works.

00:01:21.700 --> 00:01:25.040
<v Gerald Friedland>And different speakers can speak and you always sound similar to somebody else.

00:01:25.500 --> 00:01:26.240
<v Dave Jones>Right, yeah.

00:01:26.240 --> 00:01:30.040
<v Chris Gammell>It's also what everybody is used to these days.

00:01:30.860 --> 00:01:41.820
<v Chris Gammell>You know, Siri and Alexa and all these other systems that are becoming very popular are clearly speaker independent and continuous speech as well.

00:01:42.300 --> 00:01:52.800
<v Bertrand>I would expect that if Dave was speaking to the same device as I was, that it would recognize his crazy Aussie accent at the same time as my beautiful Midwest stroll.

00:01:52.800 --> 00:01:56.040
<v Chris Gammell>Well, as a matter of fact, you'd be really surprised.

00:01:56.160 --> 00:01:58.740
<v Chris Gammell>We have tried on a variety of accents in our system.

00:01:59.100 --> 00:02:01.800
<v Chris Gammell>And, you know, I have a light French accent.

00:02:01.900 --> 00:02:02.820
<v Chris Gammell>At least I'd like to think so.

00:02:03.440 --> 00:02:05.460
<v Chris Gammell>Gerald has a German accent.

00:02:05.600 --> 00:02:09.500
<v Chris Gammell>We've tried on Indian accents and all kinds of accents of English speakers.

00:02:09.560 --> 00:02:11.220
<v Chris Gammell>And it actually works really well.

00:02:11.220 --> 00:02:12.280
<v Chris Gammell>I don't know, mate.

00:02:12.280 --> 00:02:13.200
<v Dave Jones>I reckon I could break it.

00:02:13.240 --> 00:02:13.780
<v Dave Jones>No worries, mate.

00:02:13.780 --> 00:02:14.240
<v Dave Jones>She'll be right.

00:02:18.540 --> 00:02:20.460
<v Chris Gammell>Well, yeah, that might be an issue.

00:02:21.680 --> 00:02:23.220
<v Chris Gammell>There are some caveats.

00:02:23.440 --> 00:02:23.640
<v Bertrand>Yeah.

00:02:23.700 --> 00:02:30.460
<v Bertrand>Another thing we should mention here is that, so, okay, so this is the audience and you guys are running a Kickstarter and we'll get into all the details of that.

00:02:30.460 --> 00:02:37.020
<v Bertrand>But the interesting thing to me, the most interesting thing is actually not the speaker independent, but actually the network independent.

00:02:37.140 --> 00:02:39.620
<v Bertrand>The fact that, you know, I'm used to having a phone.

00:02:40.440 --> 00:02:41.160
<v Bertrand>I talk to it.

00:02:41.260 --> 00:02:46.140
<v Bertrand>I say directions to 1234 Main Street, you know, Cleveland, Ohio.

00:02:46.420 --> 00:02:47.780
<v Bertrand>And it recognizes that.

00:02:47.780 --> 00:02:57.500
<v Bertrand>But it's doing that because it's passing that audio file up to the cloud, processing it, and then sending back basically like a JSON or a text file basically to the Maps program.

00:02:57.800 --> 00:02:59.800
<v Bertrand>Then, you know, it's sending back text.

00:03:00.160 --> 00:03:05.320
<v Bertrand>You guys are able to do a localized recognition of that kind of thing.

00:03:05.420 --> 00:03:07.820
<v Bertrand>So how the hell are you doing that?

00:03:07.960 --> 00:03:14.820
<v Bertrand>I mean, how are you separate from the massive cloud that does a lot of the processing that people are used to?

00:03:14.820 --> 00:03:21.820
<v Chris Gammell>So let me just go back to this issue of cloudless, as we call it, before we go into the technical detail.

00:03:22.040 --> 00:03:22.100
<v Chris Gammell>Okay.

00:03:22.520 --> 00:03:25.440
<v Chris Gammell>You know, it's great to do things in the cloud.

00:03:25.640 --> 00:03:27.240
<v Chris Gammell>There's many reasons to go through that.

00:03:27.380 --> 00:03:33.520
<v Chris Gammell>But the reality is there's plenty of occasions where, first of all, you may not have a connection to the cloud.

00:03:33.700 --> 00:03:36.360
<v Chris Gammell>And we all know how well those Wi-Fi connections work.

00:03:37.460 --> 00:03:42.500
<v Chris Gammell>And I think the second issue, which is largely, you know, understated, is just privacy.

00:03:42.500 --> 00:03:49.680
<v Chris Gammell>I mean, we all know, you know, a lot of people don't seem to be terribly concerned, but I believe it's a real concern, privacy.

00:03:50.120 --> 00:03:53.580
<v Chris Gammell>And doing things local essentially get rid of this particular issue.

00:03:54.200 --> 00:04:01.860
<v Chris Gammell>You've got companies, I mean, a few months ago, Samsung got in trouble because their new TVs that essentially listen to voice commands,

00:04:01.860 --> 00:04:11.260
<v Chris Gammell>was using a third party and sending unencrypted text or speech, I'm sorry, to their third party for speech or text conversion,

00:04:11.800 --> 00:04:14.300
<v Chris Gammell>which basically anybody could have, you know, if dropped on.

00:04:14.940 --> 00:04:15.700
<v Gerald Friedland>Right, right, right.

00:04:15.700 --> 00:04:20.320
<v Chris Gammell>Yes, as a, you know, as a private citizen, it is of concern.

00:04:20.880 --> 00:04:24.020
<v Chris Gammell>So we really wanted a solution that was standalone.

00:04:25.480 --> 00:04:32.020
<v Chris Gammell>You know, there's nothing that goes out of that board except, hey, I recognize sentence number four that you told me to recognize.

00:04:32.340 --> 00:04:34.340
<v Chris Gammell>And that's what it does.

00:04:35.040 --> 00:04:35.720
<v Bertrand>Yeah, that's powerful.

00:04:35.960 --> 00:04:44.700
<v Dave Jones>Although I've got to say, Chris, you are a young whippersnapper because I used a voice, speaker independent voice recognition chip back in 1908, early 1980s.

00:04:44.700 --> 00:04:46.960
<v Bertrand>And we all have seen your video.

00:04:47.620 --> 00:04:49.120
<v Bertrand>Yes, you did a video about this.

00:04:49.340 --> 00:04:50.080
<v Dave Jones>You could buy the local Tandy store.

00:04:50.200 --> 00:04:54.500
<v Dave Jones>Granted, it was pretty crap and it had like 10 words, go, stop, start.

00:04:55.020 --> 00:04:55.120
<v Dave Jones>Right.

00:04:55.260 --> 00:04:56.360
<v Dave Jones>But, hey, it worked.

00:04:56.360 --> 00:04:56.660
<v Dave Jones>That's right.

00:04:56.780 --> 00:04:58.680
<v Dave Jones>And it was, you know, no, there's cloud rubbish.

00:04:59.340 --> 00:05:00.520
<v Dave Jones>Jeez, didn't even have the.

00:05:01.900 --> 00:05:02.760
<v Bertrand>Right, right.

00:05:02.960 --> 00:05:04.220
<v Bertrand>Well, yeah, and you could also say that.

00:05:04.220 --> 00:05:04.780
<v Bertrand>It was very primitive.

00:05:05.180 --> 00:05:10.260
<v Bertrand>You could say that, like, oh, well, phones talk to you today, but then you could also say, well, the speak and spell was doing this.

00:05:10.460 --> 00:05:13.280
<v Bertrand>You know, they were also, it was electronics talking to you back in the day.

00:05:13.660 --> 00:05:14.200
<v Bertrand>You know, like it's.

00:05:14.200 --> 00:05:17.140
<v Chris Gammell>Texas Instruments made a lot of money on that chip.

00:05:17.260 --> 00:05:17.760
<v Bertrand>Yeah, right.

00:05:17.760 --> 00:05:18.140
<v Bertrand>Oh, they did.

00:05:18.400 --> 00:05:18.560
<v Bertrand>Yeah.

00:05:19.140 --> 00:05:24.240
<v Bertrand>But it's just the levels of how much of that stack is really working for you, right?

00:05:24.240 --> 00:05:24.360
<v Bertrand>Yeah.

00:05:24.360 --> 00:05:25.580
<v Bertrand>It's a different experience, I think.

00:05:25.820 --> 00:05:29.020
<v Gerald Friedland>So the speak and spell is a synthesizer only, of course, right?

00:05:29.060 --> 00:05:35.280
<v Gerald Friedland>So they were not recognizing, but we were basically spelling and then we'll speak, which is an awesome toy, by the way.

00:05:35.280 --> 00:05:36.520
<v Gerald Friedland>We still have that here.

00:05:36.520 --> 00:05:43.420
<v Gerald Friedland>And I don't know where it is, but we had it like at least for, it probably still worked last year.

00:05:43.420 --> 00:05:54.820
<v Gerald Friedland>And anyway, the reason why we can make it work is because there's a combination of sort of, you know, newly available chips that make this happen.

00:05:55.400 --> 00:05:57.220
<v Gerald Friedland>And also basically the combination.

00:05:57.220 --> 00:06:01.940
<v Gerald Friedland>And I mean, I've been working in audio and multimedia for 15 years.

00:06:01.940 --> 00:06:08.260
<v Gerald Friedland>And there's basically just a lot of tricks you can do because it's not a dictation application.

00:06:08.840 --> 00:06:10.660
<v Gerald Friedland>It's basically voice commands.

00:06:10.940 --> 00:06:12.860
<v Gerald Friedland>And so we can recognize full sentences.

00:06:12.860 --> 00:06:16.740
<v Gerald Friedland>And it's also, it's, you know, speaker independent.

00:06:17.660 --> 00:06:20.280
<v Gerald Friedland>And you can program your own sentences.

00:06:20.780 --> 00:06:28.820
<v Gerald Friedland>And a lot of this is happening by basically smartly combining lots of the stuff that is out there, you know, and just making sure.

00:06:29.020 --> 00:06:30.400
<v Gerald Friedland>And then timing is an issue.

00:06:31.000 --> 00:06:32.940
<v Gerald Friedland>And there's a lot of things that just have to be right.

00:06:33.020 --> 00:06:34.000
<v Gerald Friedland>And it's a lot of tinkering.

00:06:34.540 --> 00:06:37.660
<v Gerald Friedland>But that's basically, you know, it was a lot of fun to basically put this together.

00:06:37.660 --> 00:06:41.880
<v Gerald Friedland>And as definitely as an acceptance to a challenge.

00:06:42.080 --> 00:06:49.800
<v Gerald Friedland>Because what happened here is that a lot of these companies were doing, you know, you know, Apple started and also smartphones.

00:06:50.380 --> 00:06:56.200
<v Gerald Friedland>You know, even, and I usually say as a joke, the PlayStation 4 has a cloud-based speech recognizer.

00:06:56.660 --> 00:06:58.800
<v Gerald Friedland>But the PlayStation 4 has like eight CPUs in them.

00:06:58.900 --> 00:07:04.860
<v Gerald Friedland>It's basically as powerful as some of the compute clusters that we use to do speech recognition on a research basis.

00:07:04.860 --> 00:07:06.420
<v Dave Jones>So what's the deal with that?

00:07:06.800 --> 00:07:07.400
<v Gerald Friedland>That's nuts.

00:07:07.660 --> 00:07:16.900
<v Gerald Friedland>And so they do it because then they have, you know, full control over all this training and can update the models for new words and so on.

00:07:17.300 --> 00:07:25.700
<v Gerald Friedland>But what we wanted to do is we wanted to give that full control to the maker, to the, you know, person that builds an Arduino project.

00:07:25.840 --> 00:07:29.460
<v Gerald Friedland>So that they can customize it the way they want and they have full control over it.

00:07:29.700 --> 00:07:31.060
<v Gerald Friedland>So that was the goal.

00:07:31.180 --> 00:07:35.940
<v Gerald Friedland>And it's actually, you know, again, I'm quite proud of that you can do that.

00:07:35.940 --> 00:07:39.100
<v Gerald Friedland>You can totally program your own sentences, your own call sign.

00:07:39.200 --> 00:07:40.340
<v Gerald Friedland>You can change all the timings.

00:07:40.460 --> 00:07:43.040
<v Gerald Friedland>You can change, you know, the responses, all of this.

00:07:43.620 --> 00:07:45.060
<v Dave Jones>And you can download those to the chip.

00:07:45.060 --> 00:07:47.320
<v Gerald Friedland>And you can, yes, exactly.

00:07:47.320 --> 00:07:48.900
<v Gerald Friedland>You can upload these to the board.

00:07:49.420 --> 00:07:53.480
<v Gerald Friedland>And it's training, you know, basically the part that needs to be trained on board.

00:07:53.480 --> 00:07:57.600
<v Gerald Friedland>And once you're done, you don't have to have an internet or a PC connection.

00:07:57.800 --> 00:07:59.400
<v Gerald Friedland>It's just Arduino plus the shield.

00:07:59.900 --> 00:08:02.540
<v Gerald Friedland>And you can build it into a robot, battery operated.

00:08:02.920 --> 00:08:05.940
<v Gerald Friedland>And you may be somewhere in the wild where there's actually no internet.

00:08:06.740 --> 00:08:11.840
<v Gerald Friedland>And, yeah, we've shown some of these demos on the website.

00:08:11.840 --> 00:08:18.360
<v Gerald Friedland>And also we presented it at Maker Faire, so people in the Bay Area, so Maker Faire Bay Area, they could probably just basically try it out in life.

00:08:18.540 --> 00:08:19.800
<v Bertrand>Right, which is where I saw it, actually.

00:08:20.720 --> 00:08:20.920
<v Dave Jones>Yeah.

00:08:20.920 --> 00:08:21.340
<v Dave Jones>All right.

00:08:21.380 --> 00:08:23.960
<v Dave Jones>Yes, you've just seen the demo or heard the demo.

00:08:24.440 --> 00:08:26.400
<v Dave Jones>Or no, you don't hear it.

00:08:26.400 --> 00:08:28.100
<v Dave Jones>I enabled the demo with my voice.

00:08:28.320 --> 00:08:28.660
<v Dave Jones>You controlled the demo.

00:08:28.940 --> 00:08:30.580
<v Dave Jones>My radio voice.

00:08:30.580 --> 00:08:36.600
<v Dave Jones>Please correct me if I'm wrong, but there are two types of speaker independent voice recognition.

00:08:36.600 --> 00:08:42.900
<v Dave Jones>One is the trained version, which you have to train before it can really do anything useful.

00:08:43.100 --> 00:08:48.840
<v Dave Jones>And there's the Holy Grail, which is the untrained one, right, which you can, you know, you don't have to do anything.

00:08:48.980 --> 00:09:00.480
<v Dave Jones>You can, anyone can just walk up to it and, you know, do things like the Dragon, like a lot of people will be familiar with the Dragon software, right, which has been around since the 80s or something.

00:09:00.480 --> 00:09:01.240
<v Dave Jones>That's been evolving.

00:09:01.360 --> 00:09:03.940
<v Dave Jones>That's a trained-based system, isn't it?

00:09:03.940 --> 00:09:09.720
<v Dave Jones>The more you train it, the better it gets at recognizing it and actually translating your voice into text.

00:09:10.080 --> 00:09:10.680
<v Dave Jones>How does your one?

00:09:11.540 --> 00:09:11.800
<v Gerald Friedland>Yes.

00:09:11.980 --> 00:09:23.620
<v Gerald Friedland>So I want to, first of all, and this is not to be too professorial or too scholar, but we usually call it speech recognition because the voice recognition is what we don't want to recognize.

00:09:23.800 --> 00:09:28.740
<v Gerald Friedland>We actually want to be independent of the voice because we want to recognize the words, right?

00:09:29.340 --> 00:09:29.860
<v Gerald Friedland>So, hey.

00:09:30.480 --> 00:09:35.720
<v Gerald Friedland>No, it's actually very common, including marketing people do that.

00:09:35.960 --> 00:09:40.380
<v Gerald Friedland>And in fact, we also, our Facebook page calls it Arduino voice control.

00:09:41.080 --> 00:09:41.380
<v Gerald Friedland>Right.

00:09:41.600 --> 00:09:46.620
<v Gerald Friedland>But in reality, so the scientific term is speech recognition for that reason.

00:09:47.060 --> 00:09:47.360
<v Gerald Friedland>Got it.

00:09:47.360 --> 00:09:50.720
<v Gerald Friedland>So the question about Dragon.

00:09:50.840 --> 00:09:59.440
<v Gerald Friedland>So if you have a dictation application like Dragon, this is much, much harder because you don't know whatever will be dictated, right?

00:09:59.640 --> 00:09:59.760
<v Gerald Friedland>That's right.

00:09:59.760 --> 00:10:05.020
<v Gerald Friedland>You're dictating full, maybe letters or full, like, papers even.

00:10:05.020 --> 00:10:07.200
<v Dave Jones>And it has to be 100% accurate, too.

00:10:07.520 --> 00:10:07.960
<v Gerald Friedland>Right.

00:10:08.240 --> 00:10:10.860
<v Dave Jones>Otherwise, people get frustrated and don't use it.

00:10:11.360 --> 00:10:11.520
<v Gerald Friedland>Right.

00:10:11.620 --> 00:10:17.040
<v Gerald Friedland>So it will never be 100%, but it has to be quite accurate so that you actually save work compared to typing it.

00:10:17.240 --> 00:10:17.400
<v Gerald Friedland>Correct.

00:10:17.400 --> 00:10:21.000
<v Gerald Friedland>And so that's why they do speaker adaptation.

00:10:21.000 --> 00:10:32.260
<v Gerald Friedland>So that means they have models for speech recognition in there, but they will adapt to your voice and make sure that the similarities they are matching are actually really, really similar to your own voice.

00:10:32.260 --> 00:10:37.720
<v Gerald Friedland>So the good news is that in our application case, we do voice control, right?

00:10:37.800 --> 00:10:41.120
<v Gerald Friedland>So you cannot build a full dictation application with what we do.

00:10:41.320 --> 00:10:41.420
<v Gerald Friedland>Right.

00:10:41.600 --> 00:10:41.820
<v Gerald Friedland>Got it.

00:10:41.820 --> 00:10:57.060
<v Gerald Friedland>What you can do, though, is you can give it hundreds of sentences and sort of build a Siri type of application where you basically say, okay, like, I mean, we have these examples where we have Eliza, for example, or we have an adventure game, well, an early, early 1980s adventure game.

00:10:57.320 --> 00:10:57.680
<v Gerald Friedland>Eliza.

00:10:57.880 --> 00:10:59.220
<v Dave Jones>Oh, that brings back memories.

00:10:59.360 --> 00:11:00.160
<v Bertrand>Wait, hold on.

00:11:00.260 --> 00:11:01.260
<v Bertrand>Young person here.

00:11:01.260 --> 00:11:02.280
<v Bertrand>Can we give us a hug?

00:11:02.280 --> 00:11:03.480
<v Bertrand>Young whippersnapper alert.

00:11:04.260 --> 00:11:05.000
<v Bertrand>Whippersnapper alert.

00:11:05.120 --> 00:11:06.480
<v Bertrand>Can we explain this?

00:11:06.840 --> 00:11:07.080
<v Gerald Friedland>Okay.

00:11:07.300 --> 00:11:11.300
<v Gerald Friedland>So Eliza was like, oh, my God, was it 70s or 80s?

00:11:11.500 --> 00:11:12.560
<v Gerald Friedland>No, 70s.

00:11:12.600 --> 00:11:13.180
<v Dave Jones>70s, yeah.

00:11:13.640 --> 00:11:14.240
<v Gerald Friedland>You can Wikipedia.

00:11:14.460 --> 00:11:21.740
<v Gerald Friedland>Weizenbaum basically came up with a thing where you would type natural sentences, a natural English

00:11:21.740 --> 00:11:26.320
<v Gerald Friedland>sentence, you know, and then it would react to it based on some keyword that spotted it.

00:11:26.720 --> 00:11:31.240
<v Gerald Friedland>And it would basically emulate sort of a passive therapist, you know, not a very good therapist.

00:11:31.260 --> 00:11:34.360
<v Dave Jones>Yeah, it would talk to you and answer your questions.

00:11:34.360 --> 00:11:35.060
<v Dave Jones>And right.

00:11:35.580 --> 00:11:36.680
<v Dave Jones>And Eliza, I'm depressed.

00:11:37.720 --> 00:11:38.200
<v Gerald Friedland>Right.

00:11:38.380 --> 00:11:39.160
<v Gerald Friedland>Pretty much.

00:11:39.660 --> 00:11:46.100
<v Gerald Friedland>And the funny thing was that with very little AI, really, you could almost fool people into it.

00:11:46.100 --> 00:11:46.820
<v Gerald Friedland>You could.

00:11:46.820 --> 00:11:47.560
<v Gerald Friedland>And of course, yeah.

00:11:47.560 --> 00:11:47.820
<v Gerald Friedland>Yeah.

00:11:48.060 --> 00:11:53.080
<v Gerald Friedland>And of course, since the 70s, this has evolved to, I mean, some Elizas are really amazing

00:11:53.080 --> 00:11:53.620
<v Gerald Friedland>at this point.

00:11:53.660 --> 00:11:55.740
<v Gerald Friedland>It's like, oh, my God, this is reacting to everything I say.

00:11:56.220 --> 00:11:59.740
<v Gerald Friedland>And so it was really interesting for us to say, well, I don't like the typing.

00:11:59.740 --> 00:12:02.660
<v Gerald Friedland>I actually want the real voice interaction.

00:12:02.840 --> 00:12:04.860
<v Gerald Friedland>And so I did this with Movi.

00:12:05.060 --> 00:12:08.380
<v Gerald Friedland>And there's a video on the Kickstarter page that shows that.

00:12:09.220 --> 00:12:13.220
<v Gerald Friedland>And the other one is the Hunter Bumpus game.

00:12:13.360 --> 00:12:17.760
<v Gerald Friedland>So basically, it's a little adventure where you go through caves and you're trying to find

00:12:17.760 --> 00:12:19.720
<v Gerald Friedland>this creature that is going to try to eat you.

00:12:20.120 --> 00:12:23.340
<v Gerald Friedland>And so if you run into the cave with a creature, it's eating you and it's game over.

00:12:23.340 --> 00:12:26.620
<v Gerald Friedland>But if you manage to shoot it first, then you win.

00:12:27.180 --> 00:12:32.060
<v Chris Gammell>But the real purpose of these examples really was to share interaction, real dialogue interaction

00:12:32.060 --> 00:12:35.600
<v Chris Gammell>where you recommend either keywords or full commands, essentially.

00:12:35.980 --> 00:12:37.940
<v Chris Gammell>And that's really key.

00:12:38.360 --> 00:12:42.680
<v Chris Gammell>And Eliza, I mean, was essentially 40 lines of basic back in those days.

00:12:42.680 --> 00:12:45.500
<v Dave Jones>Yeah, I learned a lot of programming basic.

00:12:45.720 --> 00:12:46.400
<v Dave Jones>Yeah, it's great.

00:12:46.480 --> 00:12:46.840
<v Chris Gammell>That's right.

00:12:46.840 --> 00:12:54.620
<v Chris Gammell>It was running into a 16K Apple computer, at least in my days, which is far less than any

00:12:54.620 --> 00:12:56.060
<v Chris Gammell>Arduino that you buy today.

00:12:56.220 --> 00:12:56.420
<v Chris Gammell>Right.

00:12:56.960 --> 00:13:02.620
<v Chris Gammell>And we just wanted to demonstrate that you could do all this just voice control using the setup.

00:13:02.860 --> 00:13:04.700
<v Chris Gammell>And it's trivial to program.

00:13:04.840 --> 00:13:05.640
<v Chris Gammell>That's really the idea.

00:13:05.700 --> 00:13:05.840
<v Chris Gammell>Right.

00:13:05.920 --> 00:13:06.840
<v Gerald Friedland>And that's the point.

00:13:06.840 --> 00:13:14.080
<v Gerald Friedland>It's like all of this can be done even if you basically pre-specify the sentences and

00:13:14.080 --> 00:13:18.200
<v Gerald Friedland>the vocabulary that is to be recognized, which helps the recognizer tremendously.

00:13:18.600 --> 00:13:18.740
<v Gerald Friedland>Right.

00:13:18.800 --> 00:13:21.040
<v Gerald Friedland>Because that's what you cannot do in a dictation application.

00:13:21.180 --> 00:13:22.100
<v Gerald Friedland>That's the big difference.

00:13:22.720 --> 00:13:22.880
<v Bertrand>Got it.

00:13:23.480 --> 00:13:23.640
<v Bertrand>Right.

00:13:23.860 --> 00:13:29.100
<v Bertrand>And it's almost like that the, you know, a lot of this stuff I think about in the sci-fi

00:13:29.100 --> 00:13:35.000
<v Bertrand>terms, you know, this is the promise of the future is voice commands, not necessarily

00:13:35.000 --> 00:13:39.880
<v Bertrand>full, you know, like at least all of the sci-fi stuff is always commands of like, okay, Earl

00:13:39.880 --> 00:13:40.400
<v Bertrand>Gray hot.

00:13:40.400 --> 00:13:40.640
<v Bertrand>Right.

00:13:40.640 --> 00:13:45.740
<v Bertrand>Like it wasn't like, well, I wanted to taste, you know, a little bit water today and I want

00:13:45.740 --> 00:13:47.400
<v Bertrand>it to be like a couple degrees warm.

00:13:47.460 --> 00:13:48.480
<v Bertrand>It was always just a command.

00:13:48.640 --> 00:13:53.500
<v Bertrand>It wasn't necessarily a full, you know, recognition type of thing.

00:13:54.020 --> 00:13:58.280
<v Chris Gammell>It wasn't like going to Starbucks and really have a full range of options when you were ordering

00:13:58.280 --> 00:13:59.220
<v Chris Gammell>your latte.

00:13:59.900 --> 00:14:02.180
<v Chris Gammell>Or even if it wasn't, it was still pretty good.

00:14:03.380 --> 00:14:03.560
<v Chris Gammell>Yeah.

00:14:03.800 --> 00:14:05.500
<v Chris Gammell>I wish I was going to tell.

00:14:05.680 --> 00:14:05.960
<v Dave Jones>Come on.

00:14:05.960 --> 00:14:08.580
<v Chris Gammell>You just gave me an idea for an application here.

00:14:08.680 --> 00:14:11.260
<v Chris Gammell>We should have the Starbucks voice control.

00:14:11.520 --> 00:14:11.780
<v Bertrand>Yeah.

00:14:11.980 --> 00:14:12.320
<v Bertrand>Well, yeah.

00:14:12.660 --> 00:14:15.580
<v Chris Gammell>Which actually would work really well.

00:14:16.000 --> 00:14:16.140
<v Chris Gammell>Right.

00:14:16.220 --> 00:14:18.360
<v Bertrand>I mean, you have all these K-Cups.

00:14:18.520 --> 00:14:21.300
<v Bertrand>You have these, there was another K-Cups thing where it was like a bartender.

00:14:21.800 --> 00:14:25.240
<v Bertrand>I mean, like all of these things, like it's not, you know, it's amazing when you think

00:14:25.240 --> 00:14:25.580
<v Bertrand>about it.

00:14:25.780 --> 00:14:29.680
<v Bertrand>When you move outside of natural language processing type stuff, really what we were really

00:14:29.680 --> 00:14:33.000
<v Bertrand>doing with, even with computers these days and apps and all these other things, it

00:14:33.000 --> 00:14:34.180
<v Bertrand>is still command based.

00:14:34.300 --> 00:14:35.860
<v Bertrand>It's still, you know, click a button.

00:14:35.980 --> 00:14:37.120
<v Bertrand>A button is a single command.

00:14:37.740 --> 00:14:42.240
<v Bertrand>And in this case of, you know, if you can recognize a single sentence, as long as it

00:14:42.240 --> 00:14:45.020
<v Bertrand>matches that sentence, it's what you want to end up with the output.

00:14:45.120 --> 00:14:47.340
<v Bertrand>It's a digital one or zero output kind of thing.

00:14:47.800 --> 00:14:50.720
<v Chris Gammell>And that was exactly the intent that we had when we started this.

00:14:50.800 --> 00:14:53.620
<v Chris Gammell>I mean, you know, you think of software programming as layers of programming.

00:14:54.280 --> 00:14:57.520
<v Chris Gammell>A lot of people don't think of hardware as a layer of hardware, but that's really what

00:14:57.520 --> 00:14:58.240
<v Chris Gammell>we're trying to do here.

00:14:58.240 --> 00:15:02.620
<v Chris Gammell>It's like, here's a board, give it a bunch of sentences, you know, and it's just going

00:15:02.620 --> 00:15:03.740
<v Chris Gammell>to recognize them for you.

00:15:03.800 --> 00:15:06.380
<v Chris Gammell>And whenever it gets one, it just flags it and you do something with it.

00:15:06.840 --> 00:15:12.880
<v Chris Gammell>It's, you know, it's, that's really the intent of this particular project.

00:15:13.540 --> 00:15:13.640
<v Dave Jones>Right.

00:15:13.760 --> 00:15:19.680
<v Dave Jones>So when, when somebody programs this with the sentence, is that just programmed to them

00:15:19.680 --> 00:15:24.340
<v Dave Jones>or is it a, or does it try and be speaker independent for that sentence?

00:15:24.860 --> 00:15:25.600
<v Dave Jones>How does that work?

00:15:25.600 --> 00:15:26.280
<v Gerald Friedland>Yeah.

00:15:26.440 --> 00:15:28.760
<v Gerald Friedland>So you program the sentence by typing it.

00:15:28.960 --> 00:15:29.960
<v Gerald Friedland>So you say in the...

00:15:29.960 --> 00:15:30.500
<v Dave Jones>Oh, okay.

00:15:30.560 --> 00:15:30.720
<v Dave Jones>Yeah.

00:15:30.720 --> 00:15:31.120
<v Dave Jones>All right.

00:15:31.300 --> 00:15:31.840
<v Dave Jones>Oh, okay.

00:15:31.840 --> 00:15:32.460
<v Dave Jones>That's clever.

00:15:32.520 --> 00:15:34.380
<v Dave Jones>Now we're getting fancy pantsy.

00:15:34.520 --> 00:15:34.740
<v Dave Jones>Okay.

00:15:34.920 --> 00:15:35.080
<v Dave Jones>Right.

00:15:35.200 --> 00:15:35.420
<v Dave Jones>Definitely.

00:15:35.420 --> 00:15:41.440
<v Gerald Friedland>So that's why, so, so that was a complication we had because, you know, English spelling and

00:15:41.440 --> 00:15:43.400
<v Gerald Friedland>English pronunciation really don't match.

00:15:43.960 --> 00:15:44.200
<v Gerald Friedland>Yeah.

00:15:44.360 --> 00:15:44.540
<v Gerald Friedland>Right.

00:15:44.720 --> 00:15:44.980
<v Gerald Friedland>So.

00:15:45.400 --> 00:15:45.700
<v Gerald Friedland>Yeah.

00:15:45.800 --> 00:15:46.820
<v Gerald Friedland>It's a pretty smart language.

00:15:47.060 --> 00:15:47.160
<v Gerald Friedland>Yeah.

00:15:47.480 --> 00:15:47.680
<v Gerald Friedland>Yeah.

00:15:47.680 --> 00:15:47.860
<v Gerald Friedland>Right.

00:15:48.460 --> 00:15:55.120
<v Gerald Friedland>So, so what you need to do there is there's, we included a two gigabyte and that's the two gigabyte

00:15:55.120 --> 00:15:55.560
<v Gerald Friedland>dictionary.

00:15:55.780 --> 00:15:56.860
<v Gerald Friedland>Many people ask us about this.

00:15:56.960 --> 00:15:57.720
<v Gerald Friedland>What does this do?

00:15:58.120 --> 00:16:02.540
<v Gerald Friedland>So the two gigabyte dictionary is actually giving you, you type the word and internally

00:16:02.540 --> 00:16:06.700
<v Gerald Friedland>it's taking the word and, and mapping it to pronunciation.

00:16:07.340 --> 00:16:12.740
<v Gerald Friedland>And this way the, the movie board knows what to expect when you say the sentence.

00:16:13.320 --> 00:16:13.720
<v Gerald Friedland>So.

00:16:13.720 --> 00:16:17.220
<v Gerald Friedland>And so that's what, that's also how it's speaker independent because it doesn't know anything

00:16:17.220 --> 00:16:18.060
<v Gerald Friedland>about the speaker anyway.

00:16:18.060 --> 00:16:18.100
<v Gerald Friedland>Okay.

00:16:18.280 --> 00:16:20.460
<v Dave Jones>That's a massive amount of data though.

00:16:20.620 --> 00:16:20.900
<v Gerald Friedland>Yeah.

00:16:21.020 --> 00:16:21.160
<v Gerald Friedland>Yeah.

00:16:21.160 --> 00:16:24.160
<v Dave Jones>How did you compile that sort of database?

00:16:24.940 --> 00:16:26.600
<v Gerald Friedland>These databases are online.

00:16:26.920 --> 00:16:28.320
<v Gerald Friedland>Oh, they're actually available.

00:16:28.580 --> 00:16:28.560
<v Speaker ?>Right.

00:16:28.680 --> 00:16:31.240
<v Dave Jones>Somebody, these have evolved over the decades, have they?

00:16:31.280 --> 00:16:31.580
<v Dave Jones>Yes.

00:16:32.000 --> 00:16:32.360
<v Gerald Friedland>Right.

00:16:32.460 --> 00:16:32.680
<v Gerald Friedland>Okay.

00:16:32.680 --> 00:16:35.880
<v Gerald Friedland>This is where you need the speech research experience.

00:16:36.440 --> 00:16:40.000
<v Bertrand>So when a user is using this board, does that mean that they're typing in something into

00:16:40.000 --> 00:16:40.780
<v Bertrand>like an online?

00:16:41.300 --> 00:16:43.320
<v Bertrand>So can you just walk us through the process here?

00:16:43.340 --> 00:16:44.400
<v Bertrand>Is it like the user types something?

00:16:44.780 --> 00:16:48.480
<v Bertrand>I mean, there's, there's the Arduino sketch that I saw on the site, but does that mean that

00:16:48.480 --> 00:16:52.320
<v Bertrand>it's going out to the cloud and actually grabbing these various pronunciations and downloading

00:16:52.320 --> 00:16:53.260
<v Bertrand>that whole thing to the board?

00:16:54.160 --> 00:16:54.520
<v Chris Gammell>No.

00:16:54.520 --> 00:16:59.500
<v Chris Gammell>So the, so the board has a, you know, the, the, the board has a four gig SD card on which

00:16:59.500 --> 00:17:01.180
<v Chris Gammell>two gig is the dictionary.

00:17:01.180 --> 00:17:06.420
<v Chris Gammell>So the, the method through which we program the board essentially, uh, if you take an Arduino

00:17:06.420 --> 00:17:11.800
<v Chris Gammell>sketch, basically you have a setup, you know, and in there you would essentially tell it,

00:17:11.920 --> 00:17:17.180
<v Chris Gammell>uh, learn, you know, sentence one, I want to turn on the light, learn sentence two, turn

00:17:17.180 --> 00:17:20.820
<v Chris Gammell>all the lights off, learn sentence four, three, four, five, et cetera.

00:17:20.820 --> 00:17:22.020
<v Chris Gammell>You can have as many as you want.

00:17:22.020 --> 00:17:27.060
<v Chris Gammell>Then you have a, you know, once you're done with a setup, basically the, the board will

00:17:27.060 --> 00:17:31.800
<v Chris Gammell>go through its own learning phase where basically it, it breaks down all these sentences into

00:17:31.800 --> 00:17:36.940
<v Chris Gammell>phonemes through the dictionary and, and, and figures out, you know, what to do with it.

00:17:37.160 --> 00:17:41.540
<v Chris Gammell>But that's completely, that dot part is completely transparent to the user and it's only done once

00:17:41.540 --> 00:17:42.960
<v Chris Gammell>if you change any of the sentences.

00:17:43.460 --> 00:17:48.740
<v Chris Gammell>And then all you have to do is in the main loop that you have, uh, you know, the main, uh,

00:17:48.740 --> 00:17:53.940
<v Chris Gammell>execution loop of the Arduino program, uh, you just pull them on a regular basis and you,

00:17:54.000 --> 00:17:57.000
<v Chris Gammell>you know, and you check to see if a sentence was recognized.

00:17:57.400 --> 00:18:01.260
<v Chris Gammell>And if it, if you did, it'll tell you, Hey, I recognize sentence number four and sentence

00:18:01.260 --> 00:18:04.480
<v Chris Gammell>number four was, you know, dim the lights 50%.

00:18:04.480 --> 00:18:04.920
<v Chris Gammell>Who knows?

00:18:05.000 --> 00:18:05.980
<v Chris Gammell>I mean, whatever you've programmed.

00:18:05.980 --> 00:18:11.080
<v Chris Gammell>And then you take that information and you go do something with it, uh, on your board.

00:18:11.420 --> 00:18:16.580
<v Chris Gammell>So there is, you know, it's, it's all, uh, you know, I mean, literally, I think we, we,

00:18:16.580 --> 00:18:20.900
<v Chris Gammell>we, we figured the absolute minimum number of line of codes you needed to add was like five.

00:18:21.440 --> 00:18:21.840
<v Chris Gammell>Yeah.

00:18:21.840 --> 00:18:23.560
<v Dave Jones>That's about as high level as you can get.

00:18:23.900 --> 00:18:25.000
<v Chris Gammell>That's about as high level.

00:18:25.420 --> 00:18:30.360
<v Chris Gammell>And, um, you know, that's just to recognize a couple of sentences basically, and then do

00:18:30.360 --> 00:18:30.980
<v Chris Gammell>something with it.

00:18:31.160 --> 00:18:31.380
<v Chris Gammell>Wow.

00:18:31.400 --> 00:18:32.360
<v Chris Gammell>Really, really simple.

00:18:32.360 --> 00:18:37.200
<v Chris Gammell>Uh, I mean, it's, it's, it's, it's extremely abstracted if you will, but it, I think it,

00:18:37.220 --> 00:18:40.100
<v Chris Gammell>it becomes extremely intuitive to a user.

00:18:40.700 --> 00:18:46.100
<v Chris Gammell>Uh, you know, and a lot of users, uh, on Arduino, I mean, you know, they're, they don't have 20

00:18:46.100 --> 00:18:48.720
<v Chris Gammell>years of programming experience and then it's certainly not speech experts.

00:18:49.140 --> 00:18:52.540
<v Chris Gammell>So, um, and I'm not a speech expert either myself.

00:18:52.860 --> 00:18:56.160
<v Chris Gammell>So I really wanted something that was actually easy for me to use.

00:18:57.080 --> 00:19:02.020
<v Dave Jones>Well, that's why even back in the eighties when we, like the speak and spell and the, uh,

00:19:02.020 --> 00:19:06.580
<v Dave Jones>famous SBO two five six, um, speech allophone chip, right.

00:19:06.580 --> 00:19:08.320
<v Dave Jones>Which would actually talk, right.

00:19:08.400 --> 00:19:12.960
<v Dave Jones>It was not voice recognition here, but you know, it was actually difficult to use that

00:19:12.960 --> 00:19:16.320
<v Dave Jones>chip because you had to know all about phenomes, right.

00:19:16.320 --> 00:19:19.760
<v Dave Jones>And you had to know how to concatenate them together to form a speech.

00:19:19.760 --> 00:19:25.040
<v Dave Jones>So that's why they came out with this companion chip that would just take serial ASCII in and

00:19:25.040 --> 00:19:29.180
<v Dave Jones>then it would figure out all the, all the allophones for you and generate the speech.

00:19:29.180 --> 00:19:29.820
<v Dave Jones>Exactly.

00:19:30.260 --> 00:19:36.220
<v Chris Gammell>And, and for us, so the, the, the board also has a built-in synthesizer as well, uh, with

00:19:36.220 --> 00:19:39.680
<v Chris Gammell>both male and female voice, although the male sounds a little bit like Stephen Hawkins.

00:19:40.060 --> 00:19:46.680
<v Chris Gammell>Uh, but the idea there, um, is, you know, there's a say command and you just type in plain English

00:19:46.680 --> 00:19:48.400
<v Chris Gammell>what you want the synthesizer to say.

00:19:48.900 --> 00:19:54.780
<v Chris Gammell>Uh, and again, we use the same dictionary to phonetize and make it, you know, go through

00:19:54.780 --> 00:19:55.020
<v Chris Gammell>sound.

00:19:55.020 --> 00:19:58.240
<v Chris Gammell>So trivial to build dialogue, basically.

00:19:58.600 --> 00:20:00.640
<v Dave Jones>I was going to say, so you almost got that for free.

00:20:00.740 --> 00:20:02.340
<v Dave Jones>You almost got the capabilities for free.

00:20:02.340 --> 00:20:02.520
<v Dave Jones>That's right.

00:20:02.860 --> 00:20:03.260
<v Dave Jones>Right.

00:20:03.720 --> 00:20:04.120
<v Dave Jones>Nice.

00:20:04.500 --> 00:20:09.040
<v Bertrand>Can you guys explain the, the phonemes a little bit more for, for uninitiated like myself,

00:20:09.220 --> 00:20:13.160
<v Bertrand>um, just like what they are and how, how that ends up affecting some of this stuff.

00:20:13.160 --> 00:20:16.500
<v Bertrand>Obviously Dave made a video about this and I'll link that in as well, but, um, just kind

00:20:16.500 --> 00:20:21.520
<v Bertrand>of how all this stuff fits together with the, the segments and breaking up speech.

00:20:21.520 --> 00:20:21.880
<v Gerald Friedland>Yes.

00:20:22.300 --> 00:20:27.200
<v Gerald Friedland>So pretty much what it is, is, uh, all words in all the languages, but let's keep it to

00:20:27.200 --> 00:20:29.160
<v Gerald Friedland>English are composed of.

00:20:29.800 --> 00:20:35.640
<v Gerald Friedland>So like we have 26, uh, um, letters letters in the alphabet.

00:20:36.180 --> 00:20:42.800
<v Gerald Friedland>Uh, we have sort of a set of, uh, phones and phonemes, uh, in each language.

00:20:42.800 --> 00:20:47.700
<v Gerald Friedland>So the letters in the alphabet, especially in English, only roughly correspond to the

00:20:47.700 --> 00:20:48.240
<v Gerald Friedland>phonemes.

00:20:48.240 --> 00:20:55.500
<v Gerald Friedland>Um, and so, uh, what you have to do is you basically, there's also there's phonetized

00:20:55.500 --> 00:20:56.420
<v Gerald Friedland>alphabet and so on.

00:20:56.500 --> 00:20:59.940
<v Gerald Friedland>So the point is phonemes are basically like the letters of the language, except they're

00:20:59.940 --> 00:21:03.300
<v Gerald Friedland>really corresponding to how we pronounce the language.

00:21:03.520 --> 00:21:06.180
<v Gerald Friedland>So like a is a phoneme, right?

00:21:06.180 --> 00:21:10.360
<v Gerald Friedland>And this, uh, is of course repeated in so many words, right?

00:21:10.420 --> 00:21:14.360
<v Gerald Friedland>So it's like some, some unit that, that is like the minimum unit that is repeated.

00:21:14.820 --> 00:21:19.320
<v Bertrand>So you mean a, a being not the letter, but the sound like a, y, if you were spelling it,

00:21:19.360 --> 00:21:19.540
<v Bertrand>right?

00:21:20.140 --> 00:21:20.520
<v Gerald Friedland>Exactly.

00:21:20.740 --> 00:21:23.400
<v Gerald Friedland>So a and a is different, right?

00:21:23.400 --> 00:21:27.320
<v Gerald Friedland>While the spelling would be made probably the same in English, you never know, but it

00:21:27.320 --> 00:21:28.240
<v Gerald Friedland>would probably be the same.

00:21:28.240 --> 00:21:28.420
<v Gerald Friedland>Yeah.

00:21:28.780 --> 00:21:29.020
<v Gerald Friedland>Yeah.

00:21:29.020 --> 00:21:33.260
<v Gerald Friedland>But these, these, these are also, you know, meaning bearing.

00:21:33.940 --> 00:21:38.380
<v Gerald Friedland>Uh, and, and so that's, so when they have me, when they bear meaning, that's when they

00:21:38.380 --> 00:21:39.300
<v Gerald Friedland>called phoneme.

00:21:39.860 --> 00:21:42.800
<v Gerald Friedland>And when they're just basically a random sound, it's called a phone.

00:21:43.020 --> 00:21:46.440
<v Gerald Friedland>So there's also, it's, it gets tricky at some point.

00:21:46.920 --> 00:21:48.560
<v Dave Jones>And it's not just a matter of joining them together.

00:21:48.640 --> 00:21:51.180
<v Dave Jones>You also have to add a little delays here and there, right?

00:21:51.560 --> 00:21:52.300
<v Dave Jones>As, as well.

00:21:52.940 --> 00:21:53.380
<v Gerald Friedland>Right.

00:21:53.380 --> 00:21:59.080
<v Gerald Friedland>So, and also then in depending on the language you're using, there is, uh, of course, like

00:21:59.080 --> 00:22:03.900
<v Gerald Friedland>sometimes it makes a difference whether the, the, the stresses on different syllables, right?

00:22:04.320 --> 00:22:07.040
<v Gerald Friedland>And stuff like this in English, we don't have that that much.

00:22:07.700 --> 00:22:11.880
<v Gerald Friedland>Um, and then of course, if you would ever do Mandarin, there's tonality in there.

00:22:11.880 --> 00:22:14.040
<v Gerald Friedland>So it, it, it becomes tricky.

00:22:14.460 --> 00:22:14.660
<v Bertrand>Oh, right.

00:22:14.680 --> 00:22:17.560
<v Bertrand>Like ups and downs and diphthongs and all that other stuff, right?

00:22:17.620 --> 00:22:18.520
<v Gerald Friedland>So, so exactly.

00:22:18.680 --> 00:22:22.760
<v Gerald Friedland>So that's why for now, by the way, people often ask this, is there, how easy it is to do

00:22:22.760 --> 00:22:23.240
<v Gerald Friedland>other languages?

00:22:23.460 --> 00:22:24.320
<v Gerald Friedland>It's not that easy.

00:22:24.380 --> 00:22:28.440
<v Gerald Friedland>Right now we keep it to English and, um, you know, if there's enough demand, we might

00:22:28.440 --> 00:22:29.640
<v Gerald Friedland>actually take on the championship.

00:22:29.900 --> 00:22:31.280
<v Bertrand>Do you guys do Esperanto?

00:22:31.660 --> 00:22:32.340
<v Bertrand>Any Esperanto?

00:22:33.220 --> 00:22:35.180
<v Bertrand>No, we don't speak, we don't speak minions.

00:22:35.360 --> 00:22:35.720
<v Chris Gammell>I'm sorry.

00:22:39.180 --> 00:22:39.540
<v Chris Gammell>Right.

00:22:40.600 --> 00:22:41.240
<v Chris Gammell>Not yet.

00:22:41.560 --> 00:22:41.860
<v Chris Gammell>Not yet.

00:22:41.980 --> 00:22:42.420
<v Chris Gammell>You'll get there.

00:22:42.640 --> 00:22:42.720
<v Chris Gammell>Yeah.

00:22:43.280 --> 00:22:43.640
<v Chris Gammell>Not yet.

00:22:43.720 --> 00:22:44.040
<v Chris Gammell>You'll get there.

00:22:44.040 --> 00:22:49.920
<v Chris Gammell>No, we, and you know, and the, the, the one thing we really tried really hard to do,

00:22:49.920 --> 00:22:57.600
<v Chris Gammell>we wanted to do a really, really good quality, you know, uh, project board, because there

00:22:57.600 --> 00:23:00.280
<v Chris Gammell>are so many things that are really just average out there.

00:23:00.420 --> 00:23:04.520
<v Chris Gammell>We really wanted to kind of up the, up the ante on, on the result.

00:23:05.180 --> 00:23:09.320
<v Chris Gammell>And as soon as we start focusing on multiple, I mean, there's, it's only two guys right now.

00:23:09.400 --> 00:23:11.560
<v Chris Gammell>It's Gerald and myself, uh, doing this.

00:23:11.560 --> 00:23:14.260
<v Chris Gammell>And we just don't have the resources to go, you know, to do more.

00:23:14.260 --> 00:23:17.680
<v Chris Gammell>Uh, if there's an interest, we'll definitely look at other languages.

00:23:17.680 --> 00:23:23.340
<v Chris Gammell>We've actually had lots of requests for, you know, German, uh, uh, there's some interest

00:23:23.340 --> 00:23:26.080
<v Chris Gammell>for Spanish and, and, and French as well.

00:23:26.360 --> 00:23:29.400
<v Chris Gammell>These are sort of the easy European languages, if you will.

00:23:29.920 --> 00:23:34.160
<v Chris Gammell>Uh, I'm sure there's plenty of opportunities for others, but again, let's do the first one

00:23:34.160 --> 00:23:34.740
<v Chris Gammell>really well.

00:23:34.740 --> 00:23:36.380
<v Chris Gammell>And then we'll do the other ones afterwards.

00:23:36.380 --> 00:23:36.820
<v Chris Gammell>Yeah.

00:23:36.820 --> 00:23:37.860
<v Bertrand>That's smart.

00:23:37.980 --> 00:23:38.160
<v Bertrand>Yeah.

00:23:38.280 --> 00:23:38.700
<v Bertrand>That's good.

00:23:38.700 --> 00:23:42.700
<v Dave Jones>Now you guys actually, we should talk about this.

00:23:42.800 --> 00:23:44.480
<v Dave Jones>It's a custom chip, right?

00:23:45.560 --> 00:23:46.220
<v Chris Gammell>No, no, no.

00:23:46.220 --> 00:23:46.820
<v Chris Gammell>It's not a custom.

00:23:46.940 --> 00:23:47.840
<v Chris Gammell>It's not a custom chip.

00:23:48.000 --> 00:23:48.860
<v Chris Gammell>It's a, no, it's not.

00:23:48.920 --> 00:23:50.220
<v Chris Gammell>It's a company called All Winner.

00:23:50.380 --> 00:23:51.260
<v Chris Gammell>It's a, it's a Chinese.

00:23:51.260 --> 00:23:51.540
<v Chris Gammell>Oh, right.

00:23:51.560 --> 00:23:57.820
<v Dave Jones>I thought you actually used them as a, as a, uh, like to actually do the custom chip for

00:23:57.820 --> 00:23:58.020
<v Dave Jones>you.

00:23:58.060 --> 00:23:58.560
<v Dave Jones>Oh, okay.

00:23:58.560 --> 00:23:59.340
<v Chris Gammell>It's a general.

00:23:59.520 --> 00:24:00.060
<v Chris Gammell>Oh no God.

00:24:00.140 --> 00:24:03.640
<v Chris Gammell>I, that, that would require a different amount of financing.

00:24:03.940 --> 00:24:07.020
<v Chris Gammell>This would be a major Kickstarter campaign there.

00:24:07.660 --> 00:24:08.100
<v Dave Jones>Exactly.

00:24:08.100 --> 00:24:09.960
<v Dave Jones>This is a million dollar Kickstarter.

00:24:09.960 --> 00:24:10.780
<v Dave Jones>I was going to ask about this.

00:24:10.880 --> 00:24:11.100
<v Dave Jones>Yes.

00:24:11.100 --> 00:24:13.240
<v Dave Jones>Like, you know, how has this come about?

00:24:13.960 --> 00:24:14.840
<v Dave Jones>Right, right, right.

00:24:14.840 --> 00:24:18.220
<v Dave Jones>So you're, so you're using this off the shelf chip from All Winner.

00:24:18.780 --> 00:24:19.180
<v Chris Gammell>Is that right?

00:24:19.180 --> 00:24:24.760
<v Chris Gammell>It's actually, that, that chip is using a lot of, of, um, tablets and portable electronics

00:24:24.760 --> 00:24:29.160
<v Chris Gammell>on, uh, well, I guess non-Apple and non-Seption products.

00:24:29.680 --> 00:24:31.040
<v Chris Gammell>And, uh, what does the chip do?

00:24:31.220 --> 00:24:34.460
<v Gerald Friedland>Well, the goodness about this chip is that it's initially designed for tablets.

00:24:34.700 --> 00:24:37.300
<v Gerald Friedland>So it has a whole CPU in there, you know?

00:24:37.300 --> 00:24:42.840
<v Gerald Friedland>So first of all, and then second, it also has audio input and audio output because tablets

00:24:42.840 --> 00:24:43.740
<v Gerald Friedland>usually have that.

00:24:43.860 --> 00:24:44.200
<v Gerald Friedland>Yeah.

00:24:44.200 --> 00:24:46.820
<v Gerald Friedland>And that makes it very, very convenient, uh, to use.

00:24:46.980 --> 00:24:46.980
<v Speaker ?>

00:24:46.980 --> 00:24:48.300
<v Chris Gammell>So it's low, it's low power.

00:24:48.440 --> 00:24:49.700
<v Chris Gammell>It's, it's fairly inexpensive.

00:24:50.520 --> 00:24:54.360
<v Chris Gammell>Uh, it's an ARM Cortex, uh, eight, eight inside.

00:24:54.360 --> 00:24:57.100
<v Chris Gammell>So it's, it's quite a, quite a powerful, uh, CPU.

00:24:57.320 --> 00:24:58.060
<v Bertrand>Oh, I see it now.

00:24:58.140 --> 00:24:58.320
<v Bertrand>Yep.

00:24:58.580 --> 00:24:58.800
<v Bertrand>Yeah.

00:24:58.800 --> 00:25:03.340
<v Bertrand>And that's the, that's the same company that, that, uh, that, that was used in the chip,

00:25:03.520 --> 00:25:08.760
<v Bertrand>the CHIP, that a $9 quote unquote board that was being, that's on Kickstarter right now

00:25:08.760 --> 00:25:09.100
<v Bertrand>as well.

00:25:09.100 --> 00:25:09.940
<v Bertrand>Is it the same one?

00:25:09.940 --> 00:25:10.360
<v Bertrand>That's right.

00:25:10.480 --> 00:25:10.700
<v Bertrand>Yeah.

00:25:10.820 --> 00:25:13.120
<v Bertrand>That, that we're having, let's not get into this.

00:25:13.220 --> 00:25:14.000
<v Bertrand>I think that we're here.

00:25:14.000 --> 00:25:14.160
<v Bertrand>No, of course, of course.

00:25:14.400 --> 00:25:14.480
<v Bertrand>Yeah.

00:25:15.180 --> 00:25:16.760
<v Bertrand>But, but it's the same company at least.

00:25:16.860 --> 00:25:18.900
<v Bertrand>And that's, and that's, that's basically how I'm trying to get it.

00:25:19.020 --> 00:25:19.120
<v Bertrand>So.

00:25:19.360 --> 00:25:19.440
<v Bertrand>Yeah.

00:25:19.480 --> 00:25:22.100
<v Chris Gammell>They're actually a major, they're actually a major player.

00:25:22.200 --> 00:25:25.260
<v Chris Gammell>They're actually one of the largest, uh, well, I don't know if I should say that,

00:25:25.300 --> 00:25:27.140
<v Chris Gammell>but largest, uh, customer of ARM.

00:25:27.980 --> 00:25:28.320
<v Bertrand>Yeah.

00:25:28.320 --> 00:25:28.780
<v Bertrand>No, no, no.

00:25:28.780 --> 00:25:31.500
<v Bertrand>I mean, yeah, it's, it's in lots of, uh, consumer level devices.

00:25:31.720 --> 00:25:32.000
<v Bertrand>I know that.

00:25:32.000 --> 00:25:34.020
<v Chris Gammell>Oh, very, very, very reputable company.

00:25:34.120 --> 00:25:34.340
<v Chris Gammell>Actually.

00:25:34.340 --> 00:25:38.960
<v Chris Gammell>It's not very well known by consumers, but, uh, people, people on the hardware know them.

00:25:39.300 --> 00:25:43.500
<v Bertrand>We had a, we had an article on, on our subreddit where we keep our links this week about, uh,

00:25:43.600 --> 00:25:47.900
<v Bertrand>cause all winter just came out with an eight or a, so you guys are using the 813, correct?

00:25:48.540 --> 00:25:48.720
<v Bertrand>Yeah.

00:25:48.720 --> 00:25:54.380
<v Bertrand>Uh, they also just recently released a 20, I think, or they announced the 20, something

00:25:54.380 --> 00:25:54.780
<v Bertrand>like that.

00:25:54.780 --> 00:25:56.940
<v Bertrand>So yeah, it's, it's very interesting to see this stuff.

00:25:56.940 --> 00:26:00.820
<v Bertrand>Cause it seems like, especially with all winter and these kinds of similar consumer level

00:26:00.820 --> 00:26:04.380
<v Bertrand>chips, these are the ones that are just moving super, super fast.

00:26:05.140 --> 00:26:10.100
<v Bertrand>Um, and I, I guess tied to that, does that end up affecting you guys because you, you know,

00:26:10.100 --> 00:26:10.880
<v Bertrand>you're using this chip.

00:26:10.900 --> 00:26:12.540
<v Bertrand>Is it going to be here in six months?

00:26:12.720 --> 00:26:14.200
<v Bertrand>I mean, or does that even matter?

00:26:14.420 --> 00:26:17.280
<v Chris Gammell>So the answer is, uh, yeah, the chip is going to be there in six months.

00:26:17.340 --> 00:26:19.680
<v Chris Gammell>Uh, we've actually had discussions about this.

00:26:19.820 --> 00:26:22.460
<v Chris Gammell>Uh, actually that particular chip is used all over the place.

00:26:22.460 --> 00:26:27.340
<v Chris Gammell>So there's, uh, it's actually made, I think by TSMC and basically as long as there's

00:26:27.340 --> 00:26:30.740
<v Chris Gammell>demand, they will keep, uh, as long as there's demand and as long as they're,

00:26:30.820 --> 00:26:32.980
<v Chris Gammell>the process is available at TSMC.

00:26:33.140 --> 00:26:34.460
<v Chris Gammell>Uh, it will get manufactured.

00:26:35.180 --> 00:26:39.780
<v Chris Gammell>Uh, you know, you might have to go, you might have to go and buy 35,000 at a time, but, uh,

00:26:39.780 --> 00:26:40.800
<v Chris Gammell>it's definitely available.

00:26:41.180 --> 00:26:45.580
<v Chris Gammell>Now the, the good, the really good news, if you look at the architecture that we're using,

00:26:45.660 --> 00:26:50.500
<v Chris Gammell>you know, the, the underlying of what we're doing is, is really a Linux, you know, we're

00:26:50.500 --> 00:26:52.180
<v Chris Gammell>running a Debian on it.

00:26:52.700 --> 00:26:55.540
<v Chris Gammell>Uh, we did a whole bunch of things to have it boot in a couple seconds.

00:26:55.540 --> 00:27:01.200
<v Chris Gammell>Uh, so really if, you know, at some point, uh, it's very certain that this chip will

00:27:01.200 --> 00:27:05.020
<v Chris Gammell>not be available, but we'll have to go and use one of its, uh, you know, future brother.

00:27:05.740 --> 00:27:10.160
<v Chris Gammell>Um, you know, it will probably end up with a redesign of the board on the next generation,

00:27:10.160 --> 00:27:13.620
<v Chris Gammell>uh, and, you know, recompile and keep going.

00:27:13.620 --> 00:27:13.780
<v Chris Gammell>Yeah.

00:27:13.960 --> 00:27:15.800
<v Bertrand>So that's not really a showstopper.

00:27:15.800 --> 00:27:18.500
<v Bertrand>That's the benefit of doing high level, like programming like that.

00:27:18.580 --> 00:27:19.660
<v Bertrand>That's great actually.

00:27:19.920 --> 00:27:20.100
<v Bertrand>Yeah.

00:27:20.400 --> 00:27:24.180
<v Chris Gammell>And it's, and it's, and it's really a hybrid of, of low level and high level.

00:27:24.180 --> 00:27:27.740
<v Chris Gammell>If you think about it, I mean, it's high level in a sense that, yeah, we are programming

00:27:27.740 --> 00:27:28.400
<v Chris Gammell>at higher level.

00:27:28.660 --> 00:27:31.200
<v Chris Gammell>It's low level in a sense that we kind of control the hardware.

00:27:31.460 --> 00:27:35.800
<v Chris Gammell>So, and we like the idea of having just that running on that board.

00:27:35.800 --> 00:27:37.680
<v Chris Gammell>Um, it helps us control timing.

00:27:37.780 --> 00:27:40.340
<v Chris Gammell>It helps us control, uh, all kinds of things.

00:27:40.340 --> 00:27:42.260
<v Chris Gammell>I mean, you're doing real time speech recognition.

00:27:42.260 --> 00:27:46.360
<v Chris Gammell>You know, you want your system to react in 300 to 500 milliseconds.

00:27:47.120 --> 00:27:51.320
<v Chris Gammell>Uh, so you don't want, you know, all of a sudden, you know, a service interrupt, you

00:27:51.320 --> 00:27:52.540
<v Chris Gammell>know, servicing your screen.

00:27:53.480 --> 00:27:53.680
<v Chris Gammell>Right.

00:27:53.820 --> 00:27:54.380
<v Chris Gammell>Or you're dizzy.

00:27:54.560 --> 00:27:56.460
<v Chris Gammell>And also, again, the system become unresponsive.

00:27:56.580 --> 00:27:58.500
<v Chris Gammell>So that's one of the advantages of controlling the hardware.

00:27:59.300 --> 00:27:59.480
<v Chris Gammell>Definitely.

00:27:59.600 --> 00:28:05.100
<v Dave Jones>Is there any, um, would your ultimate goal be a custom processor?

00:28:05.100 --> 00:28:10.760
<v Dave Jones>Like I say, there are other speech, uh, recognition, uh, voice recognition chips out

00:28:10.760 --> 00:28:14.180
<v Dave Jones>there, like the, you know, the HM2007 or whatever it is.

00:28:14.240 --> 00:28:15.820
<v Dave Jones>Um, ones like that.

00:28:15.980 --> 00:28:18.160
<v Dave Jones>Would you want to go for a custom if you could?

00:28:18.280 --> 00:28:20.640
<v Dave Jones>Is there any huge advantage to that?

00:28:20.860 --> 00:28:25.820
<v Chris Gammell>I have 25 years of, of doing chip design and custom chip designs.

00:28:26.080 --> 00:28:26.520
<v Dave Jones>Yeah.

00:28:26.600 --> 00:28:26.900
<v Dave Jones>Right.

00:28:27.180 --> 00:28:29.200
<v Chris Gammell>And the answer is absolutely not.

00:28:29.200 --> 00:28:35.800
<v Chris Gammell>Unless, unless, unless, unless you're going to be selling millions of systems.

00:28:35.940 --> 00:28:36.080
<v Chris Gammell>Right.

00:28:36.080 --> 00:28:39.000
<v Chris Gammell>So there's, there's, there's a, there's a threshold doing a custom chip of something

00:28:39.000 --> 00:28:42.200
<v Chris Gammell>like this is a multimillion dollar undertaking.

00:28:42.980 --> 00:28:48.240
<v Chris Gammell>Uh, if somebody shows one, you know, if someone shows us, uh, that there is a 50 to a hundred

00:28:48.240 --> 00:28:52.920
<v Chris Gammell>million dollar opportunity out there for doing this and is willing to, you know, uh, finance

00:28:52.920 --> 00:28:55.780
<v Chris Gammell>us, Hey, I'll, we'll gladly take their money and go and run with it.

00:28:55.780 --> 00:28:57.360
<v Chris Gammell>But we're not there today.

00:28:57.720 --> 00:29:00.780
<v Chris Gammell>And, you know, we, you know, we're taking it one stride at a time.

00:29:00.920 --> 00:29:04.040
<v Dave Jones>It just seems a shame that everything goes high level.

00:29:04.120 --> 00:29:04.440
<v Dave Jones>Oh yeah.

00:29:04.440 --> 00:29:06.000
<v Dave Jones>Let's just do it in a Linux.

00:29:06.120 --> 00:29:10.560
<v Dave Jones>Let's just do it in an ARM processor running Linux, you know, you're, you know,

00:29:10.660 --> 00:29:13.320
<v Dave Jones>Well, but there's good company there though.

00:29:13.740 --> 00:29:14.740
<v Dave Jones>I know there is.

00:29:15.080 --> 00:29:19.060
<v Bertrand>So my, my, my cell phone actually, I think it was one of the first ones doing it.

00:29:19.080 --> 00:29:24.340
<v Bertrand>Maybe not, but I know that the Moto X, I have a Moto X and it runs a coprocessor just

00:29:24.340 --> 00:29:26.620
<v Bertrand>sitting there listening for voice commands.

00:29:26.620 --> 00:29:28.480
<v Bertrand>And it's the same thing where it's a high level.

00:29:28.660 --> 00:29:31.040
<v Bertrand>It's just, they just, they don't like have a custom processor.

00:29:31.140 --> 00:29:32.480
<v Bertrand>I think they just doubled up the processor.

00:29:32.800 --> 00:29:33.120
<v Gerald Friedland>Right.

00:29:33.360 --> 00:29:34.620
<v Bertrand>I could be wrong about that, but.

00:29:34.860 --> 00:29:39.300
<v Gerald Friedland>So, so, so I, to be frank though, that, that, that is the appeal for makers too.

00:29:39.300 --> 00:29:39.580
<v Gerald Friedland>Right.

00:29:39.620 --> 00:29:44.220
<v Gerald Friedland>I mean, what I like is when I saw this first, that really reminded me of the Commodore 64

00:29:44.220 --> 00:29:49.080
<v Gerald Friedland>really, where you have, you know, your basic, you had your basic, but then you had the

00:29:49.080 --> 00:29:52.320
<v Gerald Friedland>peaks and pokes where you actually could actually directly access, access the hardware.

00:29:52.320 --> 00:29:58.520
<v Gerald Friedland>And in some way it's the same thing where right now you have GPIOs and you can actually, you

00:29:58.520 --> 00:30:00.500
<v Gerald Friedland>know, turn on and off individual lines.

00:30:00.880 --> 00:30:07.320
<v Gerald Friedland>And, and I mean, you could, some of the chip, for example, is double, two or three functions

00:30:07.880 --> 00:30:09.200
<v Gerald Friedland>on the same output pins.

00:30:09.280 --> 00:30:12.400
<v Gerald Friedland>And so you can actually switch those functions, which is kind of interesting.

00:30:12.540 --> 00:30:13.560
<v Gerald Friedland>We actually had to do this.

00:30:13.880 --> 00:30:15.320
<v Gerald Friedland>It's part of the development.

00:30:15.320 --> 00:30:20.740
<v Gerald Friedland>And, and then, and then you could create a costume kernel, you know, for in our case,

00:30:20.820 --> 00:30:25.280
<v Gerald Friedland>for example, there's no TCP stack in it because we can, don't have internet on the board again.

00:30:25.700 --> 00:30:26.680
<v Gerald Friedland>And so on and so forth.

00:30:26.780 --> 00:30:28.300
<v Gerald Friedland>So you go very, very low level.

00:30:28.440 --> 00:30:33.000
<v Gerald Friedland>And at the same time, you can run a bash script or a Python thing on it.

00:30:33.000 --> 00:30:38.380
<v Gerald Friedland>So you have the full range of, of, of capabilities there.

00:30:38.960 --> 00:30:42.500
<v Gerald Friedland>And you really, you own it, which is really interesting.

00:30:42.660 --> 00:30:47.100
<v Gerald Friedland>And as far as I know, also the ACE 13 itself allows bare metal programming.

00:30:47.360 --> 00:30:49.020
<v Gerald Friedland>So you could actually do it without Linux.

00:30:50.140 --> 00:30:50.360
<v Gerald Friedland>Right.

00:30:50.740 --> 00:30:51.320
<v Gerald Friedland>So that's killer.

00:30:51.960 --> 00:30:56.720
<v Dave Jones>But it's still not power and size optimized for the task though.

00:30:56.720 --> 00:31:00.880
<v Dave Jones>But to do that requires full custom and millions of dollars, as you said.

00:31:01.640 --> 00:31:01.780
<v Chris Gammell>Yeah.

00:31:01.900 --> 00:31:07.100
<v Chris Gammell>It's a, it's a million dollar undertaking basically that, that can be justified if you're targeting particular markets.

00:31:07.680 --> 00:31:09.820
<v Chris Gammell>And I think it's, it's going to be really interesting.

00:31:09.820 --> 00:31:15.180
<v Chris Gammell>I mean, you know, I think one of the outcome of what we, you know, what Joel and I did is we're showing it.

00:31:15.240 --> 00:31:21.960
<v Chris Gammell>You really can do it and you can actually do a pretty good product out of this that, you know, I mean, could literally be in a lot of devices.

00:31:22.680 --> 00:31:22.760
<v Chris Gammell>Right.

00:31:22.840 --> 00:31:24.360
<v Chris Gammell>Cost is, cost is an issue.

00:31:24.360 --> 00:31:33.940
<v Chris Gammell>When you look at the kind of bill of material, you know, raw bill of material, and I'm counting everything, you're talking 10, 12 dollars, you know, of, of stuff in medium quantity.

00:31:34.500 --> 00:31:37.780
<v Chris Gammell>Might be 20% cheaper, 30% cheaper if you're building millions.

00:31:37.980 --> 00:31:45.900
<v Chris Gammell>But, you know, even if it's $10, you really, you know, it's, it's hard to justify that, that additional cost to a toaster.

00:31:46.340 --> 00:31:48.320
<v Chris Gammell>Now that I won't necessarily talk to my toaster.

00:31:49.320 --> 00:31:53.220
<v Chris Gammell>But so at $10, it's going to be adequate for some applications.

00:31:53.220 --> 00:31:56.560
<v Chris Gammell>But it's going to be way too expensive for other things.

00:31:56.640 --> 00:31:57.920
<v Chris Gammell>I think that's the reality.

00:31:58.120 --> 00:32:01.600
<v Chris Gammell>So at some points, once people realize, oh my God, we can actually do this.

00:32:02.200 --> 00:32:10.620
<v Chris Gammell>You know, if there are people out there who want to come talk to us, hey, we'll gladly listen and work with them to do something like this.

00:32:10.620 --> 00:32:19.840
<v Chris Gammell>But, you know, the first step was really to show people, first, it can be done, and it can be pretty cost effective for a lot of solutions.

00:32:20.200 --> 00:32:21.820
<v Chris Gammell>And yeah, I'm sure we can make it a lot cheaper.

00:32:22.360 --> 00:32:24.100
<v Chris Gammell>But yeah, $10 is killer.

00:32:24.460 --> 00:32:28.280
<v Bertrand>I mean, that's, that's a really, I mean, like, $10 for a robotics project is nothing, right?

00:32:28.300 --> 00:32:29.180
<v Bertrand>I mean, that's, that's a server.

00:32:29.840 --> 00:32:31.220
<v Chris Gammell>You know, that's, that's the bomb.

00:32:31.320 --> 00:32:31.820
<v Chris Gammell>That's the bomb.

00:32:31.880 --> 00:32:33.720
<v Chris Gammell>That wouldn't be how much, you know, the thing is sold for.

00:32:33.720 --> 00:32:35.440
<v Chris Gammell>Oh, I know, but even still, I mean, like, yeah.

00:32:35.440 --> 00:32:36.420
<v Chris Gammell>You still have to do everything else.

00:32:36.480 --> 00:32:44.260
<v Chris Gammell>But that's, I mean, the incremental bomb, bill of material for somebody who's already doing a board and the assembly and everything else would be of that order.

00:32:44.700 --> 00:32:44.980
<v Bertrand>Right.

00:32:45.180 --> 00:32:48.040
<v Chris Gammell>So, yeah, it's, it's a pretty cheap add-on, really.

00:32:48.680 --> 00:32:48.920
<v Bertrand>Yeah.

00:32:48.920 --> 00:32:51.100
<v Dave Jones>So is this all open source?

00:32:51.640 --> 00:32:56.540
<v Dave Jones>Is this open source hardware or are you guys actually starting a company and the technology is proprietary?

00:32:57.340 --> 00:33:01.520
<v Chris Gammell>So we're actually using a lot of open source software to do all this.

00:33:01.520 --> 00:33:02.760
<v Chris Gammell>Yeah.

00:33:02.760 --> 00:33:03.140
<v Chris Gammell>Yeah.

00:33:03.140 --> 00:33:06.300
<v Dave Jones>I'm more talking about like the, the algorithms and everything else.

00:33:06.440 --> 00:33:08.260
<v Chris Gammell>So all the algorithms are open source.

00:33:08.420 --> 00:33:11.500
<v Chris Gammell>There's, I mean, there's a lot of, a lot of packages we're using.

00:33:11.600 --> 00:33:18.060
<v Chris Gammell>We're actually going to put everything on the site once, once we actually release the product, because first of all, we have to, and we want to give credit to where it's due.

00:33:18.580 --> 00:33:23.300
<v Chris Gammell>A lot of the know-how is really how you put all these things together and getting it to work.

00:33:23.800 --> 00:33:29.460
<v Chris Gammell>We're not entirely decided yet how much of it we want to, to share at this point.

00:33:29.460 --> 00:33:32.520
<v Chris Gammell>You know, and the hardware itself is fairly straightforward.

00:33:32.740 --> 00:33:35.440
<v Chris Gammell>I mean, you know, it's, it's basically a credit card.

00:33:36.180 --> 00:33:42.520
<v Chris Gammell>It's a big ass processor with a half a gig, half a gig of memory and a four gigabyte, you know, SD card.

00:33:42.700 --> 00:33:42.980
<v Chris Gammell>Okay.

00:33:43.080 --> 00:33:44.560
<v Chris Gammell>Well, there you have it.

00:33:44.900 --> 00:33:49.700
<v Chris Gammell>As I mean, you know, we're attending, we're going to a conference next week and we're showing, you know, what, what it is exactly.

00:33:49.880 --> 00:33:51.620
<v Chris Gammell>So there's really no secret there.

00:33:51.780 --> 00:33:51.920
<v Chris Gammell>Right.

00:33:52.060 --> 00:33:55.020
<v Bertrand>And that will be ESC we should mention because some of our listeners may be going.

00:33:55.020 --> 00:33:55.400
<v Bertrand>All right.

00:33:55.580 --> 00:33:55.700
<v Bertrand>Yep.

00:33:55.880 --> 00:33:56.360
<v Chris Gammell>That's right.

00:33:56.420 --> 00:33:56.540
<v Chris Gammell>Yeah.

00:33:56.540 --> 00:33:58.240
<v Chris Gammell>In Santa Clara, California.

00:34:00.020 --> 00:34:05.960
<v Chris Gammell>And so at this point, we're not, at this point, we're not open sourcing the actual flow.

00:34:06.080 --> 00:34:07.820
<v Chris Gammell>That's, that's the one limitation we're doing.

00:34:08.120 --> 00:34:08.320
<v Bertrand>Right.

00:34:08.500 --> 00:34:08.900
<v Bertrand>That's okay.

00:34:09.060 --> 00:34:16.080
<v Bertrand>I mean, like, yeah, I mean, like where you, everybody's going to make the decision about where, where you're going to make the money, where you're going to make the decision to be open or closed.

00:34:16.200 --> 00:34:17.900
<v Bertrand>It's fine with me.

00:34:17.900 --> 00:34:38.620
<v Bertrand>I mean, like you guys put a lot of work into this, so, and, and so like, so when, when you say open or close though, like, so when, when you think about this into the future and we talked about the bomb costs as well, do you envision this being always, like, do you envision licensing this out as a, you know, like buy this processor, have the software stack, you're good to go.

00:34:38.700 --> 00:34:42.060
<v Bertrand>You basically pay a license fee type thing or, or how do you envision it working?

00:34:42.060 --> 00:34:47.680
<v Chris Gammell>So, you know, the, yes, that's, that's the next step for us in some respect.

00:34:47.760 --> 00:34:54.560
<v Chris Gammell>I mean, some people, there's really two types of, of, of users in, in, at least in our experience.

00:34:54.560 --> 00:34:59.960
<v Chris Gammell>The one that just want, you know, here's the board, you know, they're, they're, I call them the system integrator, basically.

00:34:59.960 --> 00:35:03.500
<v Chris Gammell>It's like, oh, I've got this great idea and I just want to buy a board that does this.

00:35:03.520 --> 00:35:07.660
<v Chris Gammell>And they don't really care about, you know, saving money there really, if you will.

00:35:07.660 --> 00:35:19.300
<v Chris Gammell>So there's, I can, but the, obviously the other, the other model is to license essentially the stack and, you know, basically have people talk to us.

00:35:19.440 --> 00:35:24.980
<v Chris Gammell>We'll, we'll show them exactly, you know, how it's done, what it takes, how to implement it on whatever platform you have.

00:35:24.980 --> 00:35:28.140
<v Bertrand>Put this process on your board, put those other stuff and install this, right?

00:35:28.640 --> 00:35:29.040
<v Chris Gammell>Exactly.

00:35:29.280 --> 00:35:31.340
<v Chris Gammell>I mean, basically we'll be holding their hands.

00:35:31.340 --> 00:35:46.060
<v Chris Gammell>Because, you know, we, we have the expertise and, and literally in a few days we can get this up and running on, on your particular Linux distribution or on a core of your, you know, a Snapdragon from Qualcomm or something like that.

00:35:46.140 --> 00:35:48.640
<v Chris Gammell>There's plenty of ways you can, you can do that.

00:35:49.580 --> 00:35:52.680
<v Dave Jones>How much processor grunt do you actually need?

00:35:52.680 --> 00:36:00.200
<v Dave Jones>Do you actually need a big, um, Cortex I8 or whatever it is and, you know, working at gigahertz or hundreds of megahertz?

00:36:00.600 --> 00:36:02.560
<v Dave Jones>Can, can this be done like lower end?

00:36:03.040 --> 00:36:05.240
<v Dave Jones>What's actually driving the response time there?

00:36:05.240 --> 00:36:07.320
<v Bertrand>A couple op amps and like some TTL on it?

00:36:07.460 --> 00:36:08.240
<v Bertrand>Or what are we talking about?

00:36:08.240 --> 00:36:11.680
<v Gerald Friedland>So, so, so, so the answer is, the answer is yes, but at what quality?

00:36:12.240 --> 00:36:13.420
<v Gerald Friedland>And so the way I did this.

00:36:13.420 --> 00:36:14.320
<v Gerald Friedland>What are quality?

00:36:14.520 --> 00:36:15.500
<v Gerald Friedland>Someone is speaking.

00:36:15.780 --> 00:36:16.660
<v Gerald Friedland>We know this much.

00:36:16.960 --> 00:36:18.700
<v Bertrand>There is definitely sound going on.

00:36:19.680 --> 00:36:20.880
<v Bertrand>Or is it a dog barking?

00:36:20.980 --> 00:36:21.460
<v Bertrand>I'm not sure.

00:36:21.460 --> 00:36:29.980
<v Gerald Friedland>So, so, so, I mean, you can build a board that just like to, let's say, discriminate between yes and no, you know, that would, that would be not hard.

00:36:30.120 --> 00:36:33.760
<v Gerald Friedland>Maybe not even, uh, maybe even speaker independently, that will be not that hard.

00:36:34.080 --> 00:36:39.960
<v Gerald Friedland>But we do really, I mean, we can do hundreds of sentences and, um, we, we give the soft limit.

00:36:40.080 --> 00:36:42.060
<v Gerald Friedland>There are 200 sentences, which is our limit.

00:36:42.160 --> 00:36:43.720
<v Gerald Friedland>Like, this is where we stop testing.

00:36:44.360 --> 00:36:48.220
<v Gerald Friedland>Um, you know, if people manage to put on more, we have no restriction on there.

00:36:48.420 --> 00:36:51.320
<v Gerald Friedland>Um, and the major point is that we not just.

00:36:51.460 --> 00:36:53.240
<v Gerald Friedland>We have a lot of models in there.

00:36:53.240 --> 00:36:57.340
<v Gerald Friedland>So we have the, again, the acoustic model, which is basically matching these phonemes.

00:36:57.460 --> 00:37:00.020
<v Gerald Friedland>But then there's also a language model in there.

00:37:00.020 --> 00:37:16.420
<v Gerald Friedland>So basically, we, we built the language model based on the sentences so that when you recognize a sentence, um, when, when you say a sentence, um, that it's, it's, it's similar to another sentence based on the similarity.

00:37:16.420 --> 00:37:18.200
<v Gerald Friedland>You know, we can still match it, for example.

00:37:18.200 --> 00:37:23.260
<v Gerald Friedland>But then also the point is that if it doesn't get it a hundred percent, it will try to match

00:37:23.260 --> 00:37:25.040
<v Gerald Friedland>it against the sentences it has.

00:37:25.660 --> 00:37:29.340
<v Gerald Friedland>And so there's, there's this acoustic model, a language model as a matcher.

00:37:29.480 --> 00:37:32.800
<v Gerald Friedland>And if you combine all these two, then you get the quality up.

00:37:32.800 --> 00:37:38.120
<v Gerald Friedland>And of course, the more you add models and the more you add algorithms to make it better, the more you need CPU.

00:37:38.920 --> 00:37:39.440
<v Gerald Friedland>And right.

00:37:39.880 --> 00:37:42.300
<v Dave Jones>Because it's a brute force type approach.

00:37:42.300 --> 00:37:49.220
<v Dave Jones>Like if you've put in 200, uh, sentence commands in there, it's got to search all 200 until it finds a match.

00:37:49.220 --> 00:37:51.540
<v Dave Jones>Or is it more advanced than that?

00:37:51.600 --> 00:37:53.860
<v Dave Jones>Can it narrow down the options?

00:37:53.860 --> 00:37:54.180
<v Dave Jones>Yes.

00:37:54.420 --> 00:37:54.680
<v Dave Jones>Yes.

00:37:54.680 --> 00:37:55.420
<v Dave Jones>Which ones to search?

00:37:55.820 --> 00:37:57.160
<v Gerald Friedland>So there are two ways.

00:37:57.440 --> 00:38:00.180
<v Gerald Friedland>Uh, actually we give two options to the, to the programmers.

00:38:00.740 --> 00:38:03.040
<v Gerald Friedland>Um, one option is to do the raw recognition.

00:38:03.040 --> 00:38:13.740
<v Gerald Friedland>So for example, if you train it to recognize one, two, three, then, uh, you can recognize two, three, one or two, two, three or one, one, three as well.

00:38:14.000 --> 00:38:14.440
<v Gerald Friedland>Right.

00:38:14.980 --> 00:38:25.160
<v Gerald Friedland>Um, but the language model will favor two, two, three, uh, in some way, because if you, if it basically says two, two, um, uh, it sounds like three, then it's going to be two, two, three.

00:38:25.540 --> 00:38:30.340
<v Gerald Friedland>Um, and so the other thing though, is this, this is the first model we call this raw output.

00:38:30.340 --> 00:38:32.180
<v Gerald Friedland>It's very, very useful for passwords.

00:38:32.180 --> 00:38:35.360
<v Gerald Friedland>For example, if you actually want to ask for a password, you do it this way.

00:38:35.680 --> 00:38:39.380
<v Gerald Friedland>The other mode that we have is you say, no, I actually want an exact match.

00:38:39.800 --> 00:38:50.560
<v Gerald Friedland>And that means we'll, for all the sentences you put in, if you say two, two, four, and the only other sentences that has two, two, send it is two, two, three, then we'll force match it to two, two, three.

00:38:50.940 --> 00:38:53.620
<v Gerald Friedland>Um, because you say that's the closest one, right?

00:38:54.340 --> 00:38:59.780
<v Bertrand>Um, does, does this mean that the, uh, the higher likelihood of success for, for programming this as a user?

00:38:59.780 --> 00:39:08.280
<v Bertrand>Would not be to sing, have single words, but instead to have sentences because there'd just be more matching criteria against the sentence versus a single word.

00:39:08.840 --> 00:39:09.320
<v Gerald Friedland>Yes.

00:39:09.400 --> 00:39:11.240
<v Gerald Friedland>And that is the beauty of language models.

00:39:11.240 --> 00:39:18.940
<v Gerald Friedland>So in fact, you know, if you say light on, light off, you, because it's just on and off, you know, these two words are very, very similar.

00:39:19.220 --> 00:39:22.280
<v Gerald Friedland>There's really just one phoneme different differentiating it.

00:39:22.280 --> 00:39:23.400
<v Gerald Friedland>It's way harder.

00:39:23.520 --> 00:39:30.780
<v Gerald Friedland>But if you say, you know, make it bright versus I want to sleep, it's going to be, you know, basically accurate.

00:39:31.200 --> 00:39:31.320
<v Gerald Friedland>Yeah.

00:39:31.740 --> 00:39:34.340
<v Gerald Friedland>And that's, that's the beauty of language models.

00:39:34.700 --> 00:39:36.240
<v Chris Gammell>There's one thing I want to touch on.

00:39:36.360 --> 00:39:42.320
<v Chris Gammell>Uh, you were talking just earlier, cause I'm not sure we quite answered your question regarding the, the CPU power.

00:39:42.320 --> 00:39:46.820
<v Chris Gammell>There is one very important thing, which is the overall response time.

00:39:47.440 --> 00:39:53.660
<v Chris Gammell>So we don't use a hundred percent of the CPU obviously, but when we do recognition, we want a quick answer.

00:39:53.660 --> 00:40:06.420
<v Chris Gammell>So if you're actually having a dialogue with you or you're talking to the, the, to the board and you want to answer, you know, we expect a, you know, half a second response time, basically.

00:40:06.660 --> 00:40:11.140
<v Chris Gammell>You know, if you have to wait five seconds for the board to decide what it heard, it's unbearable.

00:40:11.140 --> 00:40:14.180
<v Chris Gammell>And it's, it's completely, I mean, the human interaction is, is horrible.

00:40:14.180 --> 00:40:23.660
<v Chris Gammell>So we need that amount of power to essentially get that response time, which is within a very acceptable, um, human interaction, uh, expectation.

00:40:24.020 --> 00:40:24.460
<v Bertrand>Right.

00:40:24.500 --> 00:40:27.540
<v Bertrand>If you had a, if you had a butler, you'd expect about a half second as well.

00:40:27.540 --> 00:40:27.740
<v Bertrand>Right.

00:40:28.260 --> 00:40:28.820
<v Bertrand>Yeah, exactly.

00:40:28.960 --> 00:40:34.260
<v Chris Gammell>I mean, you don't want to feel like you're talking to somebody who's on the moon and have a six second, you know, response time.

00:40:34.260 --> 00:40:41.860
<v Dave Jones>Where is at what, you know, is there a, is there something that's a real bottleneck in the software to the process in time?

00:40:41.860 --> 00:40:45.260
<v Dave Jones>Um, or is it just a whole combination of a whole bunch of things?

00:40:45.660 --> 00:40:47.920
<v Gerald Friedland>Um, it's a whole combination of a bunch of things.

00:40:48.360 --> 00:40:56.520
<v Gerald Friedland>Um, but also we, since we allow training, you know, when you pass it 200 sentences right now, you have to wait, you know, 20 to 30 seconds.

00:40:56.520 --> 00:41:05.920
<v Gerald Friedland>So that training process is, takes long, but again, the training process is only done once sort of, you know, when you change the sentences, um, basically once per project sort of, right.

00:41:06.420 --> 00:41:08.100
<v Gerald Friedland>Um, or if you update the project.

00:41:08.460 --> 00:41:11.840
<v Gerald Friedland>Um, but, um, uh, in reality.

00:41:11.860 --> 00:41:16.000
<v Gerald Friedland>It's, it's, it's, we want to have a smooth response and like everything else.

00:41:16.000 --> 00:41:22.180
<v Gerald Friedland>Like for example, if there's more noise, um, you know, you have to take care of the processing takes a little longer and so on.

00:41:22.180 --> 00:41:26.880
<v Gerald Friedland>If you want to just basically be, be able to smooth to respond, uh, to whatever comes in there.

00:41:27.880 --> 00:41:31.100
<v Dave Jones>Um, how do you deal with, um, echo and things like that?

00:41:31.100 --> 00:41:39.380
<v Dave Jones>Like if, you know, you've got an omnidirectional mic on the board and somebody's, you know, all the way on the other side of the room shouting and rooms, there's not much furniture.

00:41:39.380 --> 00:41:40.580
<v Dave Jones>It's echoing like battery.

00:41:40.580 --> 00:41:41.460
<v Bertrand>Podcast you're talking here.

00:41:41.720 --> 00:41:42.300
<v Dave Jones>We need to be easy.

00:41:44.300 --> 00:41:44.700
<v Chris Gammell>Yeah.

00:41:44.800 --> 00:41:46.060
<v Chris Gammell>You guys are audio guys.

00:41:46.120 --> 00:41:46.600
<v Chris Gammell>I can tell.

00:41:47.020 --> 00:41:49.620
<v Chris Gammell>Um, so echo is an issue.

00:41:49.740 --> 00:41:52.820
<v Chris Gammell>So what we did is actually, we, we put a lot of care.

00:41:53.000 --> 00:41:58.980
<v Chris Gammell>The, the, the, the board we're doing is more than just, uh, a computer board that we spend a lot of time on the audio front end as well.

00:41:58.980 --> 00:42:00.600
<v Chris Gammell>Uh, we want a system.

00:42:00.600 --> 00:42:08.260
<v Chris Gammell>You can actually talk from, you know, a distance that would work from literally a few inches away to about 12 feet or so.

00:42:08.260 --> 00:42:10.600
<v Chris Gammell>Um, it really depends on the room condition.

00:42:10.600 --> 00:42:12.400
<v Chris Gammell>Now, ideally you don't have echo.

00:42:12.480 --> 00:42:13.940
<v Chris Gammell>We don't do echo cancellation.

00:42:14.320 --> 00:42:17.520
<v Chris Gammell>Um, the onboard microphone is just an electric microphone.

00:42:17.680 --> 00:42:20.560
<v Chris Gammell>It's, you know, it's, it's fairly broad in terms of its pickup.

00:42:20.680 --> 00:42:21.520
<v Chris Gammell>It's omnidirectional.

00:42:22.120 --> 00:42:25.000
<v Chris Gammell>Uh, there's no, we don't, we don't do echo cancellation.

00:42:25.000 --> 00:42:30.160
<v Chris Gammell>So if somebody had an application where there is echo, one of, one of several things is going to happen.

00:42:30.280 --> 00:42:33.600
<v Chris Gammell>Either it's going to reduce the range at which they can talk considerably.

00:42:33.840 --> 00:42:36.760
<v Chris Gammell>Literally, they might have to be just a couple feet away when you talk to it.

00:42:37.000 --> 00:42:39.120
<v Chris Gammell>That's, that's, that's a real issue.

00:42:39.780 --> 00:42:46.860
<v Chris Gammell>Alternatively, we, we offer, you know, they can actually put a, uh, we have, uh, you can put an external microphone and bypass the internal one.

00:42:47.080 --> 00:42:53.140
<v Chris Gammell>Uh, and then you could put a, uh, you know, uh, an array of microphone that, that does echo cancellation basically.

00:42:53.320 --> 00:42:53.480
<v Chris Gammell>Right.

00:42:53.480 --> 00:42:56.560
<v Chris Gammell>But this becomes, this becomes very project specific.

00:42:56.900 --> 00:42:59.500
<v Chris Gammell>And so, you know, we can't solve every single problem.

00:42:59.740 --> 00:43:04.740
<v Chris Gammell>We, we can't have went for, for, you know, I would call the most common case there.

00:43:05.140 --> 00:43:06.400
<v Chris Gammell>Uh, we think it's going to work.

00:43:06.520 --> 00:43:09.540
<v Chris Gammell>It's going to actually, we know we've tried it in a lot of condition.

00:43:09.640 --> 00:43:10.420
<v Chris Gammell>It works really well.

00:43:10.940 --> 00:43:22.460
<v Chris Gammell>Uh, there's cases, for instance, where loud environment, when we were at maker fair, uh, the average, the average, yeah, we had a db meter that recorded an average of, uh, 82 db.

00:43:22.460 --> 00:43:23.680
<v Dave Jones>Oh, wow.

00:43:23.680 --> 00:43:23.980
<v Dave Jones>Constantly.

00:43:24.380 --> 00:43:28.500
<v Chris Gammell>So there was absolutely no way you could talk to it, you know, at 12 feet.

00:43:28.500 --> 00:43:33.560
<v Chris Gammell>And here we had to use, uh, a headset, but with a headset, it worked, it worked perfectly.

00:43:33.780 --> 00:43:44.420
<v Chris Gammell>I mean, actually, to tell you the truth, we were, we were a little skeptical at first, and then we're actually, uh, very pleased how well, um, the audio front end responded to that.

00:43:44.420 --> 00:43:48.020
<v Chris Gammell>And, uh, I mean, Chris, I'm not sure if you got a chance to try it there, but.

00:43:48.240 --> 00:43:48.440
<v Bertrand>I did.

00:43:48.520 --> 00:43:48.980
<v Bertrand>Yeah, yeah, yeah.

00:43:48.980 --> 00:43:49.260
<v Bertrand>I did.

00:43:49.480 --> 00:43:51.240
<v Bertrand>So can you tell us a little bit about the front end?

00:43:51.300 --> 00:43:55.660
<v Bertrand>I mean, is it all built into the silicon, uh, on the A13 or is there external op amps?

00:43:55.940 --> 00:43:56.420
<v Bertrand>No, no, it's external.

00:43:56.500 --> 00:44:02.420
<v Chris Gammell>It's external op amps, basically doing, doing good, uh, automatic gain control and, and other things as well.

00:44:02.420 --> 00:44:05.040
<v Bertrand>So what, what kind of op amps, I mean, like what, what's on board?

00:44:05.040 --> 00:44:09.400
<v Bertrand>I mean, is it just an ADC or the AGC auto control op amp?

00:44:09.600 --> 00:44:09.880
<v Bertrand>Exactly.

00:44:11.100 --> 00:44:18.560
<v Chris Gammell>We're using a good quality AGC to essentially, I mean, the idea is you want to be able to, to pick up something that's in a quiet environment.

00:44:18.560 --> 00:44:21.380
<v Chris Gammell>You want to be, you know, pick up the sound.

00:44:21.740 --> 00:44:27.880
<v Chris Gammell>So the basic principle there, you know, when it comes to audio, especially speech recognition is, you know, garbage in, garbage out.

00:44:27.880 --> 00:44:33.760
<v Chris Gammell>If, if the signal that you're getting in is, is terrible, your speech recognizer is not going to do anything, say anything good.

00:44:33.760 --> 00:44:40.280
<v Chris Gammell>So you have to get good, uh, good audio, uh, input, uh, and try to essentially normalize it.

00:44:40.460 --> 00:44:48.680
<v Chris Gammell>Um, the advantage of doing it in analog, if you will, is that, um, uh, you get a very, very fast response time.

00:44:48.760 --> 00:44:52.160
<v Chris Gammell>So if somebody starts shouting, you, you adjust, you know, immediately.

00:44:52.440 --> 00:44:59.160
<v Chris Gammell>Uh, a lot of the, uh, ADCs that, that these built-in, uh, CPUs have are fairly low quality.

00:44:59.160 --> 00:45:01.460
<v Chris Gammell>You know, they can do one or two bit adjustment.

00:45:01.460 --> 00:45:08.360
<v Chris Gammell>But it is 12 bit, it's not going to give you any, you know, if you had a 24 bit, you know, ADC in there.

00:45:08.720 --> 00:45:08.900
<v Chris Gammell>Yeah.

00:45:08.900 --> 00:45:14.620
<v Chris Gammell>You might be able to, you probably would be able to do it digitally, but that's not the type of, of, um, of what we have.

00:45:14.880 --> 00:45:19.640
<v Bertrand>So are you, are you willing to, uh, tell us what the part number is or do we have to wait for Dave to do a teardown for this?

00:45:20.020 --> 00:45:23.000
<v Chris Gammell>Uh, no, you'll have to wait till we get the board out.

00:45:23.380 --> 00:45:23.900
<v Chris Gammell>Ah, come on.

00:45:23.900 --> 00:45:24.800
<v Chris Gammell>Yeah, I know, I know.

00:45:24.940 --> 00:45:25.600
<v Bertrand>We're going to get there.

00:45:25.680 --> 00:45:26.260
<v Bertrand>We'll get there.

00:45:26.260 --> 00:45:27.800
<v Chris Gammell>Well, you'll, you'll get there.

00:45:27.860 --> 00:45:28.520
<v Chris Gammell>You'll, you'll get them.

00:45:28.580 --> 00:45:29.480
<v Chris Gammell>You'll get it soon enough.

00:45:29.600 --> 00:45:29.880
<v Bertrand>Okay.

00:45:29.920 --> 00:45:30.220
<v Bertrand>Okay.

00:45:30.940 --> 00:45:35.000
<v Bertrand>So some, some, some form of, uh, the audio AGC.

00:45:35.140 --> 00:45:35.500
<v Bertrand>Exactly.

00:45:35.960 --> 00:45:36.080
<v Bertrand>Exactly.

00:45:36.080 --> 00:45:37.300
<v Bertrand>You know, I've had bad experience.

00:45:37.440 --> 00:45:38.480
<v Bertrand>So Dave knows this as well.

00:45:38.540 --> 00:45:44.200
<v Bertrand>So we, but both Dave and I have an, uh, H1, um, uh, recorder for audio stuff.

00:45:44.920 --> 00:45:46.860
<v Bertrand>And the auto gain on that thing is terrible.

00:45:47.160 --> 00:45:52.320
<v Bertrand>I don't know if it's just the, I think it might be a digital algorithm type of thing where it's, it's so discreet.

00:45:52.320 --> 00:45:58.500
<v Bertrand>It just steps it up and it's like, it's like, you can, you like hear it when you, when you listen back to the recording, it's like, it's like, hello, this is Chris.

00:45:58.500 --> 00:46:01.780
<v Bertrand>And I'm now in a bathroom and blah, blah, blah.

00:46:01.960 --> 00:46:03.460
<v Bertrand>You know, and it's just like, it's terrible.

00:46:03.640 --> 00:46:05.300
<v Bertrand>It's just like, it steps up like that.

00:46:05.360 --> 00:46:06.680
<v Bertrand>So how do you, how do you deal with that?

00:46:06.720 --> 00:46:07.780
<v Bertrand>I mean, is it, is it gradual?

00:46:08.100 --> 00:46:08.260
<v Gerald Friedland>Yeah.

00:46:08.360 --> 00:46:12.200
<v Gerald Friedland>So the issue here was that was actually quite some work to figure this out.

00:46:12.280 --> 00:46:20.600
<v Gerald Friedland>Um, especially against the closed circuit that we don't know from A13 because A13 has a, has a microphone input.

00:46:20.600 --> 00:46:22.280
<v Gerald Friedland>And we didn't know anything about it.

00:46:22.340 --> 00:46:24.100
<v Gerald Friedland>I mean, if you find something out of work, please tell me.

00:46:24.100 --> 00:46:24.280
<v Gerald Friedland>Right.

00:46:24.340 --> 00:46:25.760
<v Gerald Friedland>You can't, you can't just go listen to it.

00:46:25.800 --> 00:46:25.920
<v Gerald Friedland>Right.

00:46:26.400 --> 00:46:33.620
<v Gerald Friedland>So I, I remember, uh, several oscilloscope sessions, um, that we had on this.

00:46:33.800 --> 00:46:42.460
<v Gerald Friedland>And we also, uh, also like trying this and see the other problem that we have, um, um, is that, so now you distort the signal.

00:46:42.460 --> 00:46:42.820
<v Gerald Friedland>Right.

00:46:43.080 --> 00:46:48.200
<v Gerald Friedland>Um, because while it sounds good to humans, speech recognizer might not pick this up so well.

00:46:48.200 --> 00:46:48.640
<v Gerald Friedland>Right.

00:46:48.860 --> 00:46:52.180
<v Gerald Friedland>So, um, we, there was a lot of testing involved.

00:46:52.420 --> 00:46:57.200
<v Gerald Friedland>Um, and this is partly why we're trying to say let's wait a little bit because we're kind of proud of the solution.

00:46:57.820 --> 00:47:00.480
<v Gerald Friedland>Um, so it's, it, that was actually hard.

00:47:00.700 --> 00:47:02.380
<v Gerald Friedland>Um, that was, that was a hard part.

00:47:02.480 --> 00:47:07.260
<v Gerald Friedland>Um, a couple of months of testing until we got this right, but it was totally worth it.

00:47:07.260 --> 00:47:18.140
<v Gerald Friedland>So, I mean, um, the, the robot example, for example, that you see on the video that we just posted with, uh, Romibo would not have been possible, uh, without, uh, that part.

00:47:18.400 --> 00:47:25.920
<v Gerald Friedland>Because, you know, when, when we have, in this case, like the microphone is inside the robot and there's a fur around it.

00:47:25.920 --> 00:47:26.320
<v Gerald Friedland>Yeah.

00:47:26.460 --> 00:47:26.980
<v Gerald Friedland>Yeah.

00:47:27.020 --> 00:47:27.260
<v Gerald Friedland>Right.

00:47:27.420 --> 00:47:29.320
<v Dave Jones>And the person that's a nice windshield.

00:47:29.660 --> 00:47:29.940
<v Dave Jones>Yeah, exactly.

00:47:30.020 --> 00:47:30.760
<v Dave Jones>It's a high-pass filter.

00:47:30.920 --> 00:47:31.520
<v Dave Jones>It really is.

00:47:31.520 --> 00:47:32.200
<v Dave Jones>It's a high-pass filter.

00:47:32.200 --> 00:47:32.620
<v Dave Jones>Yeah, that's true.

00:47:32.980 --> 00:47:33.200
<v Dave Jones>Yeah.

00:47:33.280 --> 00:47:36.380
<v Dave Jones>It's a totally, totally excellent dead cat built in, you know.

00:47:36.380 --> 00:47:36.480
<v Dave Jones>Yeah.

00:47:38.260 --> 00:47:39.800
<v Gerald Friedland>Well, that is kind of true.

00:47:39.800 --> 00:47:49.740
<v Gerald Friedland>The thing is, you don't really need a windshield when, when you're inside and you're actually, you're just sort of 12, 12 to 12, 12 feet away, pretty much, uh, what it was.

00:47:49.820 --> 00:47:52.380
<v Gerald Friedland>And just said, you know, Romibo come here and it just works.

00:47:52.380 --> 00:47:56.780
<v Gerald Friedland>And, you know, in some ways I'm, I'm always surprised myself that how well that worked.

00:47:56.780 --> 00:47:59.040
<v Gerald Friedland>Um, but it was basically the result.

00:47:59.160 --> 00:48:01.040
<v Gerald Friedland>And I, I totally texted this to Patron.

00:48:01.160 --> 00:48:07.260
<v Gerald Friedland>It was totally the result of, you know, us fiddling with it until we got a nice, uh, audio front end.

00:48:07.980 --> 00:48:08.740
<v Dave Jones>That's, yeah.

00:48:08.740 --> 00:48:12.180
<v Dave Jones>So what frequency response do you need for speech?

00:48:12.860 --> 00:48:13.160
<v Dave Jones>Yeah.

00:48:13.240 --> 00:48:13.900
<v Dave Jones>That's a good question.

00:48:14.200 --> 00:48:14.560
<v Dave Jones>Recognition.

00:48:14.680 --> 00:48:14.740
<v Dave Jones>Yeah.

00:48:14.800 --> 00:48:19.080
<v Dave Jones>Like, is it, is it, is it the classic, you know, 300 Hertz to three kilohertz phone?

00:48:19.440 --> 00:48:19.760
<v Dave Jones>No, 10K.

00:48:20.020 --> 00:48:22.680
<v Dave Jones>You need, you need those, those hard consonants, Dave.

00:48:22.740 --> 00:48:23.420
<v Gerald Friedland>Oh, tell us.

00:48:23.620 --> 00:48:23.960
<v Gerald Friedland>Come on.

00:48:23.960 --> 00:48:27.260
<v Gerald Friedland>I tell you, no, so no, I mean, uh, yeah.

00:48:27.280 --> 00:48:30.540
<v Gerald Friedland>So that was basically telephone speech a long time ago.

00:48:30.980 --> 00:48:40.620
<v Gerald Friedland>Um, the three Hertz to, uh, no, what we have is, uh, we usually, for, first of all, we sample at full rate, which is 40, 40, 44, 100 or so.

00:48:40.620 --> 00:48:41.760
<v Gerald Friedland>Basically city rate.

00:48:41.940 --> 00:48:43.360
<v Gerald Friedland>Because we can, right.

00:48:43.460 --> 00:48:44.020
<v Gerald Friedland>Because we can.

00:48:44.020 --> 00:48:44.220
<v Gerald Friedland>Right.

00:48:44.300 --> 00:48:44.740
<v Dave Jones>Of course.

00:48:44.860 --> 00:48:45.060
<v Dave Jones>Right.

00:48:45.240 --> 00:48:48.680
<v Dave Jones>But we're talking about what is the minimum requirement.

00:48:48.680 --> 00:48:49.160
<v Gerald Friedland>Yeah.

00:48:49.240 --> 00:48:53.640
<v Gerald Friedland>All the models and everything is usually in speech recognition is now trained at 16 kilohertz sample rate.

00:48:53.960 --> 00:49:04.520
<v Gerald Friedland>That gives you, that gives you like a, sort of a 50 to, to eight kilohertz, uh, you know, sampling, um, uh, frequencies that you basically can pick up.

00:49:04.700 --> 00:49:10.720
<v Gerald Friedland>And you need those higher frequencies for S for distinguishing of S and T and so on and F, you know.

00:49:10.960 --> 00:49:12.820
<v Gerald Friedland>And actually F is so important.

00:49:13.020 --> 00:49:15.280
<v Gerald Friedland>Turn off, you know, you really need that.

00:49:15.280 --> 00:49:15.460
<v Gerald Friedland>Right.

00:49:16.100 --> 00:49:17.460
<v Gerald Friedland>The trailing stuff, right?

00:49:17.780 --> 00:49:18.160
<v Bertrand>Or start.

00:49:18.180 --> 00:49:18.680
<v Bertrand>It's very important.

00:49:19.280 --> 00:49:19.680
<v Bertrand>Start.

00:49:19.860 --> 00:49:21.200
<v Bertrand>Instead of start, it's start.

00:49:21.360 --> 00:49:21.520
<v Bertrand>Yeah.

00:49:21.580 --> 00:49:21.800
<v Bertrand>Right?

00:49:22.000 --> 00:49:24.040
<v Dave Jones>And for guys like me with a high pitch voice.

00:49:24.400 --> 00:49:24.700
<v Dave Jones>Yeah.

00:49:25.440 --> 00:49:26.260
<v Dave Jones>Well, yeah.

00:49:26.880 --> 00:49:27.100
<v Speaker ?>Right.

00:49:27.640 --> 00:49:27.840
<v Chris Gammell>Yeah.

00:49:27.880 --> 00:49:30.380
<v Chris Gammell>So actually you're bringing pitch, which is really interesting.

00:49:30.380 --> 00:49:36.300
<v Chris Gammell>Um, so the models, the models were, the models we're using for training are actually all adults.

00:49:36.640 --> 00:49:38.200
<v Chris Gammell>I mean, we're trained by adults.

00:49:38.760 --> 00:49:47.260
<v Chris Gammell>And so what, what we've, um, realized, and it's interesting because, you know, we're trying, we're targeting at least the maker community and there's a lot of kids there.

00:49:47.460 --> 00:50:00.360
<v Chris Gammell>Uh, and it's a, it's a great, uh, it's the, the board is really great, uh, you know, telling boys ages because essentially, um, if they haven't hit puberty, it's, it doesn't do anything.

00:50:00.360 --> 00:50:01.900
<v Chris Gammell>It doesn't do anything very well for, for boys.

00:50:02.020 --> 00:50:07.000
<v Chris Gammell>They really have to get that lower voice, uh, to really get things to work.

00:50:07.100 --> 00:50:10.820
<v Chris Gammell>So it seems to not work very well for boys under the age of 12.

00:50:10.960 --> 00:50:13.160
<v Chris Gammell>Uh, after, after 12, it works right.

00:50:13.280 --> 00:50:16.600
<v Chris Gammell>But for girls, it works at much younger ages.

00:50:16.960 --> 00:50:22.820
<v Chris Gammell>So girls usually eight years old and older seems to work without a problem.

00:50:23.000 --> 00:50:23.980
<v Chris Gammell>It's very interesting.

00:50:23.980 --> 00:50:29.420
<v Chris Gammell>And, and it's just, and it's not that, that the, uh, it really has to do with training more than anything else.

00:50:29.420 --> 00:50:33.380
<v Chris Gammell>Because the, the sampling is adult voices for all kinds of legal reasons.

00:50:33.500 --> 00:50:40.900
<v Bertrand>So one thing I also wanted to ask about is the, uh, so you mentioned response time previously, and you said about 300 to 500 milliseconds.

00:50:41.440 --> 00:50:43.500
<v Bertrand>What, what actually is that stack up?

00:50:43.540 --> 00:50:45.880
<v Bertrand>Can you tell us about the, the, you know, okay.

00:50:45.880 --> 00:50:55.120
<v Bertrand>So from the time when that five, 50 Hertz to 16 kilohertz hits the, hits the microphone, goes to the AGC, very minimal delays from there.

00:50:55.240 --> 00:50:56.900
<v Bertrand>It gets into the ADC.

00:50:57.280 --> 00:50:58.480
<v Bertrand>What happens then?

00:50:58.480 --> 00:51:00.180
<v Bertrand>Like all the way up to the top level.

00:51:00.940 --> 00:51:02.240
<v Bertrand>Um, well, yeah.

00:51:02.240 --> 00:51:09.500
<v Gerald Friedland>So, I mean, basically we, the, the kernel of the operating system sends us the sound data and packets, right?

00:51:09.600 --> 00:51:09.800
<v Gerald Friedland>Okay.

00:51:09.800 --> 00:51:09.980
<v Gerald Friedland>So.

00:51:10.440 --> 00:51:16.480
<v Dave Jones>Oh, so, so is it as, as, is, is it as the person is speaking or does it wait until a pause at the end of a sentence?

00:51:16.500 --> 00:51:17.520
<v Dave Jones>No, as the person is speaking.

00:51:17.780 --> 00:51:19.620
<v Gerald Friedland>No, it's sending packets constantly, right?

00:51:19.720 --> 00:51:21.780
<v Gerald Friedland>So you have to practice packets all the time.

00:51:22.440 --> 00:51:23.020
<v Gerald Friedland>And then, yeah.

00:51:23.120 --> 00:51:25.500
<v Gerald Friedland>And then the first thing you need to figure out is the person speaking.

00:51:25.500 --> 00:51:25.940
<v Gerald Friedland>Okay.

00:51:26.060 --> 00:51:37.000
<v Gerald Friedland>Um, so, um, in, in our case, what we'll do is we do this energy based and there is a, a fixed threshold built in, but you can also modify that threshold.

00:51:37.680 --> 00:51:38.120
<v Gerald Friedland>Okay.

00:51:38.580 --> 00:51:47.360
<v Gerald Friedland>And, and it's actually, uh, it's sort of tries to be a little intelligent because we know we have a model in there that knows how language typically works.

00:51:47.360 --> 00:51:52.840
<v Gerald Friedland>So we, for example, know that if it's something super short, like it won't be a word, right?

00:51:52.980 --> 00:51:53.660
<v Gerald Friedland>Did you hear that actually?

00:51:54.200 --> 00:51:54.360
<v Gerald Friedland>Yeah.

00:51:54.400 --> 00:51:54.820
<v Gerald Friedland>Okay, good.

00:51:55.160 --> 00:51:55.400
<v Gerald Friedland>Okay.

00:51:55.500 --> 00:52:02.660
<v Gerald Friedland>Um, or, or if it's like, uh, ongoing, never changing for 10 seconds, that's not a word either.

00:52:02.660 --> 00:52:02.960
<v Gerald Friedland>Right.

00:52:03.000 --> 00:52:05.380
<v Gerald Friedland>So there's some, there's some intelligence in there.

00:52:05.580 --> 00:52:07.820
<v Gerald Friedland>Actually, that's yet another model that needs to be calculated.

00:52:08.260 --> 00:52:09.500
<v Gerald Friedland>Usually don't even mention that.

00:52:09.740 --> 00:52:10.800
<v Bertrand>So can you, can you give things?

00:52:10.880 --> 00:52:20.540
<v Bertrand>So like an example be like someone saying no, like currently saying no, or someone saying like, like screaming, like, ah, like, is that like kind of the difference you're talking about or what?

00:52:20.780 --> 00:52:22.140
<v Gerald Friedland>That's the difference I'm talking about.

00:52:22.220 --> 00:52:22.480
<v Gerald Friedland>Yes.

00:52:22.480 --> 00:52:32.520
<v Gerald Friedland>So, because usually words are like sort of half second or a good of a, like half a second long to like two seconds long.

00:52:32.520 --> 00:52:32.780
<v Gerald Friedland>Right.

00:52:32.780 --> 00:52:33.940
<v Gerald Friedland>Depending on the long word.

00:52:33.940 --> 00:52:36.300
<v Gerald Friedland>Germans have very long words, but we're English.

00:52:36.300 --> 00:52:36.580
<v Gerald Friedland>Yeah.

00:52:36.580 --> 00:52:36.860
<v Gerald Friedland>Right.

00:52:36.860 --> 00:52:41.720
<v Gerald Friedland>Um, and, and then also there's, there's some short pauses and then there's longer pauses.

00:52:41.720 --> 00:52:46.260
<v Gerald Friedland>And these longer pauses really usually indicate somebody's done with speaking.

00:52:46.580 --> 00:52:47.020
<v Bertrand>Okay.

00:52:47.020 --> 00:52:50.360
<v Gerald Friedland>And, and so you have to catch up to typically how dialogue works.

00:52:51.020 --> 00:52:59.980
<v Gerald Friedland>Um, and, um, this is also something we can do in contrast to when you have systems that are trying to transcribe stuff because we talk to the machine.

00:52:59.980 --> 00:53:04.740
<v Gerald Friedland>And, and in fact, there's always a little bit of an adaptation going on with the human towards the machine.

00:53:05.380 --> 00:53:06.520
<v Gerald Friedland>Um, but this works pretty well.

00:53:06.600 --> 00:53:09.640
<v Gerald Friedland>I mean, you know, you see all the examples that, you know, they work with that.

00:53:09.640 --> 00:53:18.240
<v Gerald Friedland>And so once we have figured out, okay, so that's likely some speech, um, we'll basically, uh, need to recognize this.

00:53:18.320 --> 00:53:21.480
<v Gerald Friedland>And in order to recognize this, we, we, uh, extract features.

00:53:22.240 --> 00:53:28.680
<v Gerald Friedland>And then we use acoustic model to try all combinations of phonemes that this spectral envelope could represent.

00:53:29.760 --> 00:53:29.880
<v Dave Jones>Right.

00:53:29.940 --> 00:53:32.400
<v Dave Jones>I was going to say, is this done in a spectral form?

00:53:32.500 --> 00:53:36.920
<v Dave Jones>Do you like FFT this and then you look at the spectral density in each frequency band?

00:53:37.020 --> 00:53:37.620
<v Dave Jones>How does that work?

00:53:37.800 --> 00:53:38.560
<v Gerald Friedland>Sort of.

00:53:38.560 --> 00:53:42.560
<v Gerald Friedland>So what we use is what everybody uses these days is MFCC features.

00:53:42.700 --> 00:53:46.520
<v Gerald Friedland>MFCC features stand for mal frequency capsule coefficients.

00:53:47.420 --> 00:53:52.340
<v Gerald Friedland>And capstrom is an interesting word that is actually like spectrum, except not, right?

00:53:52.400 --> 00:53:53.800
<v Gerald Friedland>This is kind of a word, right?

00:53:54.100 --> 00:53:56.800
<v Gerald Friedland>And if you look that up, actually, that's, that's very interesting.

00:53:56.920 --> 00:53:57.940
<v Gerald Friedland>That's exactly what that is.

00:53:58.000 --> 00:54:04.600
<v Gerald Friedland>So what it is, is, um, you do, uh, uh, basically you take the logarithm, then you do an, uh, DCT.

00:54:04.600 --> 00:54:08.360
<v Gerald Friedland>And then you do, um, uh, what's called a mal spacing.

00:54:08.360 --> 00:54:12.380
<v Gerald Friedland>That means you do, you weight the frequencies according to typical hearing.

00:54:12.380 --> 00:54:17.140
<v Gerald Friedland>So like lower frequencies, uh, are perceived differently than higher frequencies.

00:54:17.140 --> 00:54:19.060
<v Gerald Friedland>And so you basically do this weighting.

00:54:19.060 --> 00:54:21.380
<v Gerald Friedland>And then, and that's the interesting part.

00:54:21.520 --> 00:54:23.680
<v Gerald Friedland>You do a DCT again.

00:54:24.020 --> 00:54:28.500
<v Gerald Friedland>So you take a DCT off the DCT and this is a total hack, right?

00:54:28.720 --> 00:54:33.160
<v Bertrand>So DCT, we should also mention, sorry, we should also mention DCT is district cosine transform,

00:54:33.340 --> 00:54:37.340
<v Bertrand>which is similar to an FFT, basically binning frequencies, right?

00:54:37.340 --> 00:54:37.840
<v Gerald Friedland>Yes.

00:54:37.840 --> 00:54:41.140
<v Gerald Friedland>Then the more important part is that you don't come up with complex space.

00:54:41.140 --> 00:54:45.900
<v Gerald Friedland>If you have an FFT, you have a complex values and non-complex and real values, and you don't

00:54:45.900 --> 00:54:47.800
<v Gerald Friedland>want the complex values because it's just so much data.

00:54:47.880 --> 00:54:48.900
<v Gerald Friedland>We want to do data reduction.

00:54:49.780 --> 00:54:51.280
<v Gerald Friedland>Um, so you do that.

00:54:51.280 --> 00:54:56.260
<v Gerald Friedland>And what you end up with is, is basically an interesting hack.

00:54:56.460 --> 00:55:01.180
<v Gerald Friedland>So the reason I, what you actually want to do is you want to look at this, at the spectrum,

00:55:01.180 --> 00:55:05.700
<v Gerald Friedland>but the reason we use Capstrom is because it is even less data.

00:55:05.700 --> 00:55:09.520
<v Gerald Friedland>And you basically look at, so what are the peaks in the spectrum, right?

00:55:09.940 --> 00:55:15.700
<v Gerald Friedland>Because basically now you're fitting, uh, again, cosine curves to, to the, uh, to the

00:55:15.700 --> 00:55:16.060
<v Gerald Friedland>spectrum.

00:55:16.060 --> 00:55:19.100
<v Gerald Friedland>So you find out what are the peaks in, in the spectrum.

00:55:19.280 --> 00:55:21.260
<v Gerald Friedland>And so you look at way less values.

00:55:21.260 --> 00:55:24.460
<v Gerald Friedland>And I don't want to explain this trick here because it's really old.

00:55:24.460 --> 00:55:26.880
<v Gerald Friedland>It's like, I think 60s or something.

00:55:27.220 --> 00:55:29.020
<v Gerald Friedland>And that's what, what everybody does.

00:55:29.020 --> 00:55:34.100
<v Gerald Friedland>And as an academic, I could tell you how bad it is, but the point is, it's what, it

00:55:34.100 --> 00:55:34.480
<v Gerald Friedland>works.

00:55:34.480 --> 00:55:36.920
<v Gerald Friedland>It works and it's what everybody does.

00:55:37.640 --> 00:55:42.440
<v Gerald Friedland>And so I'm, I'm, while we've, we've been trying and I have publications on this to replace

00:55:42.440 --> 00:55:43.120
<v Gerald Friedland>those features.

00:55:43.720 --> 00:55:46.020
<v Gerald Friedland>Um, you know, it's just work.

00:55:46.080 --> 00:55:47.800
<v Gerald Friedland>It's really just what everybody does.

00:55:48.400 --> 00:55:49.020
<v Gerald Friedland>And so you have these.

00:55:49.160 --> 00:55:49.980
<v Gerald Friedland>We don't judge you here.

00:55:50.640 --> 00:55:52.720
<v Bertrand>Dave and I use things that work all the time.

00:55:53.680 --> 00:55:55.120
<v Gerald Friedland>So that's basically the point.

00:55:55.180 --> 00:56:00.120
<v Gerald Friedland>So you have these features, MFCCs, and then with those features, usually you match them

00:56:00.120 --> 00:56:01.500
<v Gerald Friedland>against these phoneme models.

00:56:01.500 --> 00:56:07.080
<v Gerald Friedland>So you try to find out your sort of, you know, this, this capsule feature, it could that be

00:56:07.080 --> 00:56:08.260
<v Gerald Friedland>that phoneme or that phoneme.

00:56:08.580 --> 00:56:13.340
<v Gerald Friedland>And so you come up with a whole combination of phonemes for that particular region that

00:56:13.340 --> 00:56:14.340
<v Gerald Friedland>had high energy.

00:56:14.340 --> 00:56:19.040
<v Gerald Friedland>And then you take the language model and say, okay, which of these combinations is the most

00:56:19.040 --> 00:56:19.360
<v Gerald Friedland>likely?

00:56:19.520 --> 00:56:22.000
<v Gerald Friedland>Because we know the sentences that are most likely in there.

00:56:22.300 --> 00:56:22.660
<v Bertrand>Yeah.

00:56:22.660 --> 00:56:27.140
<v Bertrand>It's almost like a paint by, at least in my head, the way it seems like is like you already

00:56:27.140 --> 00:56:28.080
<v Bertrand>programmed in.

00:56:28.080 --> 00:56:28.640
<v Bertrand>All right.

00:56:28.660 --> 00:56:31.640
<v Bertrand>The, the, the sentences, Chris is better than Dave.

00:56:32.040 --> 00:56:35.300
<v Bertrand>That's, you know, that's, that's a set of 20 phonemes.

00:56:35.300 --> 00:56:38.860
<v Bertrand>Like it's like 20 or it's, there's like, you know, and they're all coded, right?

00:56:38.880 --> 00:56:46.240
<v Bertrand>Each of the phonemes is coded as, you know, A, F, G, H, I, E, whatever it is.

00:56:46.240 --> 00:56:46.480
<v Bertrand>Right.

00:56:46.520 --> 00:56:48.100
<v Bertrand>And then you go and match against that model.

00:56:48.140 --> 00:56:49.080
<v Bertrand>Is that, is that sound right?

00:56:49.440 --> 00:56:49.620
<v Gerald Friedland>Yeah.

00:56:49.680 --> 00:56:51.480
<v Gerald Friedland>That, that's pretty much sound right.

00:56:51.520 --> 00:56:52.280
<v Gerald Friedland>So that's the point.

00:56:52.280 --> 00:56:57.840
<v Gerald Friedland>And then, and then, um, you, you, you come up with all these sentences and then, so we do

00:56:57.840 --> 00:57:00.980
<v Gerald Friedland>this on a phoneme level and come up with words, right?

00:57:00.980 --> 00:57:05.240
<v Gerald Friedland>So as I said before, you could have one, two, three as a training.

00:57:05.240 --> 00:57:08.340
<v Gerald Friedland>And if you said two, two, three, we would actually recognize two, two, three.

00:57:08.720 --> 00:57:10.060
<v Gerald Friedland>So that's the first output.

00:57:10.160 --> 00:57:13.140
<v Gerald Friedland>And then the second output would be to match two to three.

00:57:13.360 --> 00:57:17.740
<v Gerald Friedland>If you basically ask for that to forcibly match it to one of the sentences that you were

00:57:17.740 --> 00:57:18.220
<v Gerald Friedland>trained in.

00:57:18.800 --> 00:57:20.100
<v Gerald Friedland>So that's the whole process.

00:57:20.440 --> 00:57:20.900
<v Dave Jones>Got it.

00:57:20.900 --> 00:57:27.640
<v Dave Jones>So would it, because you're continuously, um, listening, like there's no,

00:57:27.640 --> 00:57:31.240
<v Dave Jones>like a button that you push and go, right, I'm going to recognize, I'm sure your job

00:57:31.240 --> 00:57:35.300
<v Dave Jones>would be much easier if you got the user to push a button and then speak and then know

00:57:35.300 --> 00:57:37.180
<v Dave Jones>that's the start of a thing.

00:57:37.300 --> 00:57:41.040
<v Dave Jones>What is your, um, false detection rate?

00:57:41.080 --> 00:57:44.800
<v Dave Jones>Like have you programmed in, you know, a few dozen sentences and then just left it running

00:57:44.800 --> 00:57:45.900
<v Dave Jones>all day to see.

00:57:45.900 --> 00:57:49.300
<v Dave Jones>And in like in an office where people are talking all the time to see if there's any

00:57:49.300 --> 00:57:50.680
<v Dave Jones>false output.

00:57:51.100 --> 00:57:51.280
<v Gerald Friedland>Yes.

00:57:51.440 --> 00:57:53.580
<v Gerald Friedland>So, um, there's a couple things to say here.

00:57:53.740 --> 00:57:56.980
<v Gerald Friedland>Um, first of all, you can do push to talk if you want to.

00:57:56.980 --> 00:58:02.380
<v Gerald Friedland>So we, we program the API in such a way that you can either call a call, uh, put in a call

00:58:02.380 --> 00:58:05.340
<v Gerald Friedland>sign or you can put in no call sign.

00:58:05.740 --> 00:58:10.880
<v Gerald Friedland>Um, then it will react to any noise, you know, as soon as you raise your voice, it will say,

00:58:10.880 --> 00:58:11.880
<v Gerald Friedland>Oh, that's, that's me.

00:58:12.300 --> 00:58:14.780
<v Gerald Friedland>Or you can, you can not do this at all.

00:58:14.780 --> 00:58:19.600
<v Gerald Friedland>And you can basically, uh, just say, don't do any recognition at all unless I tell you.

00:58:19.600 --> 00:58:20.600
<v Gerald Friedland>Right.

00:58:20.600 --> 00:58:25.300
<v Gerald Friedland>And, and that means that if you install a button to the Arduino board and then you basically

00:58:25.300 --> 00:58:29.300
<v Gerald Friedland>question that button and the button is on, you say, okay, now, now do recognition.

00:58:29.300 --> 00:58:30.860
<v Gerald Friedland>So you can actually do a push to talk.

00:58:30.960 --> 00:58:32.740
<v Gerald Friedland>But, but, but the call sign is.

00:58:32.860 --> 00:58:35.860
<v Bertrand>Well, in your example, you also show like the videos that you guys have on the Kickstarter

00:58:35.860 --> 00:58:40.940
<v Bertrand>page, you have like a thing that says computer or, you know, what if.

00:58:41.080 --> 00:58:41.160
<v Bertrand>Yeah.

00:58:41.160 --> 00:58:45.220
<v Dave Jones>That's the thing I was going to say, like the, like the Siri, like you say Siri, because

00:58:45.220 --> 00:58:46.720
<v Dave Jones>that's the trigger keyword.

00:58:46.840 --> 00:58:48.560
<v Dave Jones>So you don't have to search everything else.

00:58:48.620 --> 00:58:49.500
<v Chris Gammell>It's exactly that.

00:58:49.580 --> 00:58:54.620
<v Chris Gammell>So the call sign really helps you get the false triggering down to hardly anything.

00:58:55.120 --> 00:58:58.340
<v Chris Gammell>It's actually, so in our case, the call sign is programmable.

00:58:58.580 --> 00:59:03.820
<v Chris Gammell>Uh, you, you can pretty much use anything, but there are some sort of, if you wish guidelines

00:59:03.820 --> 00:59:09.580
<v Chris Gammell>and the guidelines there typically is, uh, try to avoid like the five most, you know, 500

00:59:09.580 --> 00:59:14.000
<v Chris Gammell>most common English words because otherwise it'll get triggered by just standard conversation

00:59:14.000 --> 00:59:16.220
<v Chris Gammell>because it's actually really good at picking up the call sign.

00:59:16.860 --> 00:59:19.640
<v Chris Gammell>Um, typically three syllables tend to work really well.

00:59:19.640 --> 00:59:25.700
<v Chris Gammell>So actually Siri, when you use it, it's actually, Hey Siri or Amazon use Alexa, which are words,

00:59:25.700 --> 00:59:28.540
<v Chris Gammell>you know, that are really nowhere near close to anything else.

00:59:28.540 --> 00:59:29.020
<v Chris Gammell>Right.

00:59:29.180 --> 00:59:31.800
<v Chris Gammell>Uh, when we did the robot, um, uh,

00:59:31.800 --> 00:59:34.340
<v Dave Jones>So you can go like, Hey stupid robot.

00:59:34.600 --> 00:59:35.740
<v Chris Gammell>Well, so actually though.

00:59:35.800 --> 00:59:35.960
<v Chris Gammell>Yeah.

00:59:35.960 --> 00:59:41.560
<v Chris Gammell>With the robot example, we actually just posted, uh, it's called Romibo.

00:59:41.760 --> 00:59:46.060
<v Chris Gammell>So we actually program Romibo in there and Romibo really doesn't sound like anything else.

00:59:46.540 --> 00:59:49.460
<v Chris Gammell>And it picks it up just every single time.

00:59:49.460 --> 00:59:50.880
<v Chris Gammell>It's, it's, it works really well.

00:59:51.000 --> 00:59:54.320
<v Bertrand>What if someone was reciting Shakespeare and they're like, Romibo, Romibo,

00:59:54.780 --> 00:59:56.400
<v Bertrand>where's the Ratho Romibo?

00:59:56.900 --> 00:59:57.860
<v Gerald Friedland>You screwed.

00:59:58.520 --> 00:59:59.780
<v Gerald Friedland>Then you have that problem.

01:00:00.240 --> 01:00:03.960
<v Gerald Friedland>So, so, and that's why, so in terms of the second question with measuring the accuracy

01:00:03.960 --> 01:00:10.320
<v Gerald Friedland>is so, I mean, you can definitely get into, uh, human accuracies with the system.

01:00:10.600 --> 01:00:13.200
<v Gerald Friedland>Um, of course though, you can also screw it up.

01:00:13.240 --> 01:00:14.700
<v Gerald Friedland>I mean, that's the point of being a maker.

01:00:15.140 --> 01:00:16.100
<v Gerald Friedland>It can fail.

01:00:16.700 --> 01:00:22.940
<v Gerald Friedland>Um, so, I mean, again, if you make the call sign like, uh, two or a, or some of these words

01:00:22.940 --> 01:00:24.580
<v Gerald Friedland>that we use all the time, then guess what?

01:00:24.580 --> 01:00:25.880
<v Gerald Friedland>It's going to react to it all the time.

01:00:25.880 --> 01:00:30.000
<v Gerald Friedland>Um, and also if you basically program it with 200 sentences, but they're all distinguished

01:00:30.000 --> 01:00:31.800
<v Gerald Friedland>in one syllable.

01:00:32.600 --> 01:00:33.600
<v Gerald Friedland>Well, good luck.

01:00:33.740 --> 01:00:34.360
<v Gerald Friedland>I mean, that's just fine.

01:00:34.360 --> 01:00:35.460
<v Bertrand>I command that you blank.

01:00:35.680 --> 01:00:36.940
<v Bertrand>I command that you blank.

01:00:37.080 --> 01:00:38.220
<v Bertrand>I command that you blank.

01:00:38.500 --> 01:00:38.820
<v Gerald Friedland>Exactly.

01:00:39.120 --> 01:00:39.560
<v Gerald Friedland>Exactly.

01:00:39.760 --> 01:00:40.160
<v Gerald Friedland>Exactly.

01:00:40.320 --> 01:00:40.720
<v Gerald Friedland>That's it.

01:00:40.840 --> 01:00:44.980
<v Gerald Friedland>And the point is though, I challenge you then if you do that and you complain to actually

01:00:44.980 --> 01:00:48.460
<v Gerald Friedland>try that with a human because a human actually is not a hundred percent either.

01:00:48.960 --> 01:00:49.120
<v Dave Jones>No.

01:00:49.640 --> 01:00:49.820
<v Dave Jones>No.

01:00:50.080 --> 01:00:50.340
<v Gerald Friedland>Right.

01:00:50.340 --> 01:00:53.980
<v Gerald Friedland>And, and, and of course the typical things make it worse.

01:00:54.080 --> 01:00:56.620
<v Gerald Friedland>If you have heavy accent, you know, it makes it worse.

01:00:56.740 --> 01:00:59.300
<v Gerald Friedland>And if you have a lot of noise in the room, it makes it worse.

01:01:00.480 --> 01:01:01.480
<v Gerald Friedland>Um, you know.

01:01:01.580 --> 01:01:05.360
<v Dave Jones>Or if people aren't paying attention properly, they're not, you know, they're thinking about

01:01:05.360 --> 01:01:06.360
<v Dave Jones>the next thing they're going to say.

01:01:06.360 --> 01:01:11.080
<v Dave Jones>Like we often have that here on the amp hour because we're, you know, constantly trying

01:01:11.080 --> 01:01:15.580
<v Dave Jones>to communicate to people and we're trying to, you know, we might miss things and, you

01:01:15.580 --> 01:01:16.740
<v Dave Jones>know, because we're thinking about some question we're going to ask next.

01:01:16.740 --> 01:01:18.740
<v Bertrand>Dave, I have never missed a sandwich before.

01:01:18.740 --> 01:01:21.180
<v Bertrand>That's what you said, right?

01:01:21.240 --> 01:01:21.560
<v Bertrand>Sandwich?

01:01:21.680 --> 01:01:22.340
<v Bertrand>It was something about sandwiches?

01:01:23.580 --> 01:01:23.900
<v Dave Jones>Sorry?

01:01:24.240 --> 01:01:24.380
<v Dave Jones>What?

01:01:24.560 --> 01:01:24.880
<v Dave Jones>What's that?

01:01:24.920 --> 01:01:26.900
<v Dave Jones>Chris, I was too busy thinking about what I was going to talk about next.

01:01:26.900 --> 01:01:27.220
<v Dave Jones>Yeah, exactly.

01:01:27.560 --> 01:01:27.880
<v Bertrand>Sandwiches.

01:01:30.240 --> 01:01:31.980
<v Bertrand>So I was actually, I was wondering about this.

01:01:32.240 --> 01:01:35.320
<v Bertrand>So what about like, so, so obviously it's sitting there monitoring all the time.

01:01:35.660 --> 01:01:36.560
<v Bertrand>What kind of like power consumption?

01:01:36.600 --> 01:01:37.400
<v Bertrand>I don't really know.

01:01:37.480 --> 01:01:41.060
<v Bertrand>I don't know much about the A13 itself, but what kind of power consumption is it just

01:01:41.060 --> 01:01:42.420
<v Bertrand>to sit there monitoring and listening?

01:01:42.500 --> 01:01:44.520
<v Bertrand>Because I would think that's a, there's no sleep mode.

01:01:44.520 --> 01:01:45.640
<v Bertrand>It sounds very active.

01:01:45.760 --> 01:01:48.640
<v Bertrand>Even if you are just listening for a certain keyword or call sign.

01:01:48.740 --> 01:01:49.440
<v Bertrand>So, so.

01:01:49.540 --> 01:01:49.920
<v Gerald Friedland>What does it take?

01:01:50.120 --> 01:01:53.040
<v Gerald Friedland>We, we usually say it's three Watts power consumption.

01:01:53.460 --> 01:01:53.640
<v Gerald Friedland>In general.

01:01:53.760 --> 01:01:54.240
<v Dave Jones>Three Watts.

01:01:54.280 --> 01:01:56.060
<v Dave Jones>You can fly to the moon on three, three Watts.

01:01:56.060 --> 01:01:56.420
<v Dave Jones>Yes.

01:01:56.740 --> 01:01:57.060
<v Gerald Friedland>But.

01:01:57.360 --> 01:01:59.560
<v Gerald Friedland>But can you, can you tell the rocket to go there, Dave?

01:01:59.560 --> 01:01:59.820
<v Gerald Friedland>No.

01:01:59.960 --> 01:02:00.240
<v Gerald Friedland>Yeah.

01:02:00.400 --> 01:02:00.840
<v Gerald Friedland>Well, no.

01:02:00.840 --> 01:02:06.040
<v Gerald Friedland>But the point is that it is in fact taking less energy if, if it's just listening versus,

01:02:06.040 --> 01:02:07.400
<v Gerald Friedland>for example, if you're training it.

01:02:08.880 --> 01:02:14.420
<v Gerald Friedland>And we actually could actually, that's something we actually might do just actually release an

01:02:14.420 --> 01:02:17.780
<v Gerald Friedland>energy profile of, of this thing, which we haven't done.

01:02:17.780 --> 01:02:21.820
<v Gerald Friedland>But we have internally looked at it and it's actually interesting.

01:02:21.820 --> 01:02:27.300
<v Gerald Friedland>It's something that I learned in this project is that it, you can actually, you know, usually

01:02:27.300 --> 01:02:32.100
<v Gerald Friedland>you know, when you study computer science, we tell students to like, you know, runtime is

01:02:32.100 --> 01:02:32.480
<v Gerald Friedland>important.

01:02:32.640 --> 01:02:37.220
<v Gerald Friedland>Well, runtime is important, but when you do small devices, energy consumption is also important.

01:02:37.380 --> 01:02:42.160
<v Gerald Friedland>So you can, based on the algorithm you're implementing, the, the, the energy consumption changes.

01:02:42.160 --> 01:02:46.440
<v Gerald Friedland>And so we actually took this into account a little bit when we created this.

01:02:46.960 --> 01:02:49.060
<v Gerald Friedland>And so actually that's something to do.

01:02:49.160 --> 01:02:53.540
<v Gerald Friedland>It's a good, good hint for us to actually release the energy profile for it.

01:02:53.880 --> 01:02:54.280
<v Gerald Friedland>One, one.

01:02:54.480 --> 01:02:58.740
<v Dave Jones>Can people slow it down and trade off response time versus power consumption?

01:03:00.100 --> 01:03:01.860
<v Chris Gammell>It's a good question.

01:03:02.280 --> 01:03:02.620
<v Chris Gammell>Good question.

01:03:02.700 --> 01:03:03.120
<v Chris Gammell>Not yet.

01:03:03.300 --> 01:03:07.080
<v Chris Gammell>No, and this is not the way, this is probably not the way you want it.

01:03:07.100 --> 01:03:07.940
<v Chris Gammell>You would want to do it.

01:03:08.300 --> 01:03:09.440
<v Chris Gammell>It's a stretch goal.

01:03:09.720 --> 01:03:10.620
<v Chris Gammell>It's a stretch goal.

01:03:10.620 --> 01:03:10.600
<v Speaker ?>That's right.

01:03:10.600 --> 01:03:13.960
<v Bertrand>If we hit $50,000 on Kickstarter.

01:03:14.340 --> 01:03:15.820
<v Chris Gammell>We will have the power consumption.

01:03:16.100 --> 01:03:19.480
<v Chris Gammell>No, but there, I, I agree with you, Dave.

01:03:19.560 --> 01:03:23.920
<v Chris Gammell>There's definitely applications where, I mean, if let's say you wanted something like this

01:03:23.920 --> 01:03:28.020
<v Chris Gammell>that's going to be battery operated and you want to change the battery once a year type

01:03:28.020 --> 01:03:32.680
<v Chris Gammell>of thing, then essentially you'd have to, you know, essentially you would, you know,

01:03:32.720 --> 01:03:35.800
<v Chris Gammell>stop the CPU, put it in deep sleep, if you will.

01:03:35.980 --> 01:03:40.240
<v Chris Gammell>And then you just keep your analog front end and whenever you get the right energy.

01:03:40.240 --> 01:03:44.900
<v Chris Gammell>You wake up immediately, you do some processing and you shut back down.

01:03:45.060 --> 01:03:46.840
<v Chris Gammell>So these are techniques that can be done.

01:03:47.080 --> 01:03:51.060
<v Chris Gammell>We haven't, you know, this is another level in terms of complexity.

01:03:51.060 --> 01:03:52.340
<v Chris Gammell>We haven't gotten there yet.

01:03:52.980 --> 01:03:57.500
<v Dave Jones>I was going to say, is there, could you implement like a two processor solution where you have

01:03:57.500 --> 01:04:04.300
<v Dave Jones>a smaller, lower power processor just hooked onto the front end, which recognizes your keyword

01:04:04.300 --> 01:04:11.960
<v Dave Jones>and then it wakes up the heavy duty processor to process the, you know, main sentence and

01:04:11.960 --> 01:04:12.400
<v Dave Jones>things like that?

01:04:12.400 --> 01:04:13.240
<v Dave Jones>No, there's no advantage to it.

01:04:13.240 --> 01:04:16.820
<v Chris Gammell>Because essentially you'd be duplicating, you'd be duplicating everything twice.

01:04:16.960 --> 01:04:17.540
<v Chris Gammell>You'd be duplicating the effort.

01:04:17.540 --> 01:04:17.760
<v Chris Gammell>Right.

01:04:18.100 --> 01:04:21.600
<v Chris Gammell>I think just looking at energy level and whenever you get a spike in thing.

01:04:21.600 --> 01:04:23.940
<v Dave Jones>Unless you had a fixed keyword, right?

01:04:23.940 --> 01:04:29.300
<v Dave Jones>Maybe if you had like a fixed keyword, you could program that into a low power micro.

01:04:29.540 --> 01:04:29.640
<v Dave Jones>Possibly.

01:04:29.640 --> 01:04:35.560
<v Gerald Friedland>I think the reason, the real way to do this quickly is push to talk, right?

01:04:35.680 --> 01:04:36.220
<v Gerald Friedland>So you basically.

01:04:36.380 --> 01:04:36.840
<v Dave Jones>Of course.

01:04:36.960 --> 01:04:38.160
<v Dave Jones>Oh, that solves everything.

01:04:38.160 --> 01:04:38.980
<v Dave Jones>Yeah, exactly.

01:04:39.760 --> 01:04:40.480
<v Dave Jones>Yeah, right.

01:04:40.680 --> 01:04:40.860
<v Dave Jones>Right.

01:04:42.540 --> 01:04:45.060
<v Chris Gammell>But I mean, these processors have sleep modes.

01:04:45.420 --> 01:04:47.740
<v Chris Gammell>So I'm sure there are ways of doing it.

01:04:48.560 --> 01:04:53.240
<v Chris Gammell>You know, we've been in the process of solving a lot of other problems so far.

01:04:53.240 --> 01:04:53.760
<v Chris Gammell>Yeah, that's a lazy problem.

01:04:53.760 --> 01:04:56.740
<v Chris Gammell>I think that this will be sort of the next generation afterwards.

01:04:57.220 --> 01:05:03.960
<v Chris Gammell>Or if someone comes with a particular application, we'll be glad to work with them and do that

01:05:03.960 --> 01:05:04.440
<v Chris Gammell>specifically.

01:05:04.440 --> 01:05:11.480
<v Chris Gammell>But I think what's interesting is, you know, once people realize, oh my God, you can actually

01:05:11.480 --> 01:05:17.820
<v Chris Gammell>do this on something that's the size of a credit card, you know, then you start thinking

01:05:17.820 --> 01:05:19.720
<v Chris Gammell>about all the other things you can do with it.

01:05:19.980 --> 01:05:23.520
<v Chris Gammell>And, you know, that's, I think, what makes it really, really interesting.

01:05:23.660 --> 01:05:30.020
<v Chris Gammell>And for us, it's, you know, we're finishing the first leg of the journey and we have a lot

01:05:30.020 --> 01:05:31.860
<v Chris Gammell>of ideas afterwards.

01:05:31.860 --> 01:05:33.500
<v Chris Gammell>Well, that's okay.

01:05:34.120 --> 01:05:37.440
<v Dave Jones>So you guys are aware, you've reached your goal, right?

01:05:37.640 --> 01:05:37.900
<v Dave Jones>Yes.

01:05:38.180 --> 01:05:42.500
<v Dave Jones>But you're, everyone's, of course, secretly hoping, is there like a secret goal that you're

01:05:42.500 --> 01:05:42.940
<v Dave Jones>hoping for?

01:05:43.020 --> 01:05:47.220
<v Dave Jones>Oh, like we really, we asked for 12 grand, but we really need 50 otherwise, you know.

01:05:47.360 --> 01:05:47.700
<v Dave Jones>No.

01:05:47.940 --> 01:05:52.980
<v Chris Gammell>So otherwise, you know, that's, so I'm going to tell it as it is at $12,000.

01:05:52.980 --> 01:05:55.640
<v Chris Gammell>We will be losing money in the process.

01:05:56.200 --> 01:05:56.480
<v Dave Jones>Okay.

01:05:56.560 --> 01:05:56.780
<v Dave Jones>Yeah.

01:05:56.900 --> 01:05:59.980
<v Chris Gammell>We really would like, we really would like to be at 20,000.

01:06:00.240 --> 01:06:07.180
<v Chris Gammell>You know, I mean, to us, this is, this really was, the goal of the Kickstarter campaign was

01:06:07.180 --> 01:06:11.360
<v Chris Gammell>really to engage the maker community to see what the response was.

01:06:11.460 --> 01:06:16.960
<v Chris Gammell>We've been, we've been really getting a lot of really good feedback, questions from people.

01:06:16.960 --> 01:06:22.220
<v Chris Gammell>I mean, we had someone come back to us and say, Hey, my, my cousin is paraplegic.

01:06:22.320 --> 01:06:27.520
<v Chris Gammell>He can only make sounds, you know, with this work, you know, and sadly enough, the answer

01:06:27.520 --> 01:06:30.900
<v Chris Gammell>is probably not, but we actually have been working on sound recognition.

01:06:30.900 --> 01:06:35.320
<v Chris Gammell>So it's actually a problem that we know well, and there might be a way for people like this,

01:06:35.400 --> 01:06:39.260
<v Chris Gammell>but this is, this is for us a way to engage the community and make sure that we come up

01:06:39.260 --> 01:06:43.220
<v Chris Gammell>with something that, you know, people think, first of all, Hey, it's pretty cool.

01:06:43.220 --> 01:06:47.500
<v Chris Gammell>And I could actually use this, uh, that, you know, and we want to engage them.

01:06:47.620 --> 01:06:51.460
<v Chris Gammell>So part of it is, you know, engaging the community, part of this marketing, getting our name out,

01:06:51.540 --> 01:06:55.080
<v Chris Gammell>uh, you know, getting to people to realize that this kind of stuff can be done.

01:06:55.680 --> 01:06:58.880
<v Chris Gammell>Uh, and now you can start thinking about, you know, what you're going to do with it.

01:06:58.880 --> 01:07:00.740
<v Chris Gammell>We're not in there to make, you know, end product.

01:07:00.860 --> 01:07:05.260
<v Chris Gammell>I mean, this is, and this is always a difficult part of Kickstarter campaigns, which is not,

01:07:05.260 --> 01:07:08.540
<v Chris Gammell>you know, an end product where people can just buy it and put them around, right.

01:07:08.640 --> 01:07:12.240
<v Chris Gammell>You know, we're, we're selling a board, which in itself will do nothing unless you're a maker

01:07:12.240 --> 01:07:13.500
<v Chris Gammell>and you're actually going to do something with it.

01:07:13.920 --> 01:07:17.960
<v Chris Gammell>One thing I want to mention by the way, is that, you know, we picked Arduino as a platform

01:07:17.960 --> 01:07:19.600
<v Chris Gammell>because there's millions of them out there.

01:07:19.640 --> 01:07:25.780
<v Chris Gammell>It's a great platform, but the basic way the Arduino talks to the shield is actually via an

01:07:25.780 --> 01:07:26.880
<v Chris Gammell>RS 232 interface.

01:07:26.880 --> 01:07:31.480
<v Chris Gammell>So for people who actually know a little bit of that serial communications, and I'm sure

01:07:31.480 --> 01:07:34.920
<v Chris Gammell>there's a lot of people out there, there's no reason why you can just use the board and,

01:07:34.920 --> 01:07:36.320
<v Chris Gammell>you know, talk to any other system.

01:07:36.480 --> 01:07:40.360
<v Chris Gammell>So you could connect to Raspberry Pi, you can connect to your PC for like here, or these

01:07:40.360 --> 01:07:41.660
<v Chris Gammell>the RS 232 is not as common.

01:07:42.140 --> 01:07:47.520
<v Chris Gammell>Uh, but it's, it's very easy to use on other, uh, platform and we will be releasing kind of

01:07:47.520 --> 01:07:53.860
<v Chris Gammell>the low level, um, serial, uh, commands, uh, you know, uh, to be able to do all these

01:07:53.860 --> 01:07:55.460
<v Chris Gammell>things at a low level as well.

01:07:55.480 --> 01:07:56.520
<v Chris Gammell>I said the library you're saying, right?

01:07:56.640 --> 01:07:57.260
<v Chris Gammell>So right now you're in the library.

01:07:57.260 --> 01:08:00.800
<v Chris Gammell>Yeah, the library provides a really nice way, you know, a higher level, uh, you know, to

01:08:00.800 --> 01:08:04.520
<v Chris Gammell>interact with the board, but you know, all the low level commands will be, will be made,

01:08:04.680 --> 01:08:05.580
<v Chris Gammell>uh, will make that available.

01:08:05.580 --> 01:08:08.920
<v Chris Gammell>So people can really, I mean, I don't want to call it hack it with hack the board because

01:08:08.920 --> 01:08:11.940
<v Chris Gammell>you're not going to be able to go change the software inside, but you'll be able to do

01:08:11.940 --> 01:08:15.300
<v Chris Gammell>some, some very cool stuff with it that you may not necessarily be able to do with the,

01:08:15.300 --> 01:08:15.840
<v Chris Gammell>the higher level.

01:08:15.840 --> 01:08:16.240
<v Chris Gammell>Yes.

01:08:16.900 --> 01:08:18.480
<v Bertrand>As one last question about software.

01:08:18.680 --> 01:08:22.640
<v Bertrand>So, I mean, like, so all this stuff, I mean, you, you guys had mentioned, let me just look

01:08:22.640 --> 01:08:23.340
<v Bertrand>at my notes here.

01:08:23.680 --> 01:08:29.140
<v Bertrand>You had mentioned, uh, this MFCC, but like, what is all this written?

01:08:29.220 --> 01:08:32.420
<v Bertrand>I mean, like, so the, and you said, this is all in a Debian system, but like, is this

01:08:32.420 --> 01:08:36.560
<v Bertrand>all Python based or what is, what is the main languages this stuff is written in?

01:08:36.580 --> 01:08:37.800
<v Bertrand>Is it all custom or what?

01:08:38.160 --> 01:08:43.100
<v Gerald Friedland>I, you know, I like this question because I get this question all the time and I'm, I always

01:08:43.100 --> 01:08:47.480
<v Gerald Friedland>tell also, you know, to my students, I tell them, look, a real system is composed of whatever

01:08:47.480 --> 01:08:48.400
<v Gerald Friedland>you can take.

01:08:48.400 --> 01:08:48.600
<v Gerald Friedland>Right.

01:08:48.660 --> 01:08:52.720
<v Gerald Friedland>So there's like when people buy a robot at the first, it's sort of MacGyverism really.

01:08:53.480 --> 01:08:54.040
<v Gerald Friedland>Um, yeah, yeah.

01:08:54.040 --> 01:08:56.120
<v Bertrand>It's made of hot glue and popsicle sticks.

01:08:56.240 --> 01:09:00.940
<v Gerald Friedland>I mean, when you, when you build a robot, you have to take whatever part fits and this

01:09:00.940 --> 01:09:01.620
<v Gerald Friedland>is the same way.

01:09:01.620 --> 01:09:08.120
<v Gerald Friedland>So, um, this is, uh, there is, uh, you know, again, he's going to say it's Fortran, Dave.

01:09:08.340 --> 01:09:08.740
<v Gerald Friedland>No.

01:09:10.320 --> 01:09:10.960
<v Gerald Friedland>Okay.

01:09:11.220 --> 01:09:11.360
<v Gerald Friedland>Good.

01:09:11.600 --> 01:09:11.780
<v Gerald Friedland>No.

01:09:11.780 --> 01:09:16.340
<v Gerald Friedland>That is about the only language that is not in there.

01:09:17.420 --> 01:09:22.720
<v Gerald Friedland>But it's, it's made mostly, uh, uh, C for all its thing that's low level.

01:09:22.900 --> 01:09:24.780
<v Gerald Friedland>Again, there's a costume kernel and so on.

01:09:24.840 --> 01:09:26.840
<v Gerald Friedland>So I needed to go into sound drivers and so on.

01:09:27.280 --> 01:09:32.400
<v Gerald Friedland>And then, um, there's a lot of Python to glue things together.

01:09:32.400 --> 01:09:35.080
<v Gerald Friedland>Um, which is usually a good way to glue things together.

01:09:35.080 --> 01:09:35.600
<v Gerald Friedland>Yeah, definitely.

01:09:35.780 --> 01:09:35.920
<v Gerald Friedland>Yeah.

01:09:36.160 --> 01:09:38.500
<v Gerald Friedland>And, and there's a lot of shell scripting too.

01:09:38.500 --> 01:09:43.440
<v Gerald Friedland>Um, so the other thing is, for example, a board like this needs to start up, uh, a little

01:09:43.440 --> 01:09:46.620
<v Gerald Friedland>faster than let's say, you know, your typical Linux boot of a minute.

01:09:47.020 --> 01:09:51.780
<v Gerald Friedland>Um, so you need to do a lot of scripting and optimize scripts, uh, in this regard.

01:09:51.780 --> 01:09:58.500
<v Gerald Friedland>Um, so there's from basically, honestly, from very low level C programming, uh, up to, uh,

01:09:58.500 --> 01:10:01.820
<v Gerald Friedland>very high level shell scripting and Python and C and C plus plus.

01:10:01.820 --> 01:10:05.600
<v Gerald Friedland>And of course, you know, you have your Arduino code that I, that you create a C plus plus for

01:10:05.600 --> 01:10:05.980
<v Gerald Friedland>library.

01:10:05.980 --> 01:10:06.180
<v Gerald Friedland>Yeah.

01:10:06.180 --> 01:10:08.960
<v Gerald Friedland>And it's sort of this Arduino ish dialect for Arduino.

01:10:09.240 --> 01:10:10.260
<v Gerald Friedland>So it's all in there.

01:10:10.420 --> 01:10:14.300
<v Gerald Friedland>It's, it's, it's, uh, it's multi-language, multi-cultural.

01:10:14.300 --> 01:10:18.940
<v Bertrand>So, so, but what you're mostly saying is that the, the user won't really have access to nor

01:10:18.940 --> 01:10:23.780
<v Bertrand>will they need to have, uh, this kind of access to the, to the lower level programming stuff

01:10:23.780 --> 01:10:26.040
<v Bertrand>or even the high level programming stuff on, on board.

01:10:26.140 --> 01:10:30.620
<v Bertrand>It's kind of just a standalone solution and you kind of throw it chunk commands through

01:10:30.620 --> 01:10:31.360
<v Bertrand>RSDW2.

01:10:31.420 --> 01:10:32.880
<v Gerald Friedland>Well, so, I mean, basically I'm saying two things.

01:10:32.880 --> 01:10:36.220
<v Gerald Friedland>First of all, all the Arduino stuff is going to be completely open because all the Arduino

01:10:36.220 --> 01:10:37.220
<v Gerald Friedland>stuff is open anyway.

01:10:37.460 --> 01:10:38.840
<v Gerald Friedland>So on the Arduino side.

01:10:39.420 --> 01:10:45.000
<v Gerald Friedland>Um, but then also again, all the open source stuff is open because, uh, not only do we have

01:10:45.000 --> 01:10:50.580
<v Gerald Friedland>to do that, but it also is absolutely sort of in, uh, my intention to always keep open

01:10:50.580 --> 01:10:51.380
<v Gerald Friedland>what needs to be open.

01:10:51.640 --> 01:10:56.100
<v Gerald Friedland>But what I'm saying though, is it's not that easy for somebody to go just in there and change

01:10:56.100 --> 01:11:01.980
<v Gerald Friedland>it because it's such a pottery of, I mean, the real work here is to integrate it in a smart

01:11:01.980 --> 01:11:02.300
<v Gerald Friedland>way.

01:11:02.300 --> 01:11:02.740
<v Gerald Friedland>Right.

01:11:02.740 --> 01:11:04.820
<v Bertrand>It's interdependencies and stuff, right?

01:11:04.940 --> 01:11:05.160
<v Gerald Friedland>Right.

01:11:05.220 --> 01:11:05.860
<v Gerald Friedland>Interdependencies.

01:11:06.080 --> 01:11:10.680
<v Gerald Friedland>So if you look into it, it's not easy to, it's not like, actually that's the thing that

01:11:10.680 --> 01:11:14.700
<v Gerald Friedland>it might be open source, but it might not be helpful to be open source because you're like

01:11:14.700 --> 01:11:18.280
<v Gerald Friedland>in there, it's like, Oh, I need to know how to configure a speech organizer, a Linux kernel

01:11:18.280 --> 01:11:19.500
<v Gerald Friedland>at this and this and this and this.

01:11:19.840 --> 01:11:20.720
<v Gerald Friedland>And there's a lot of work.

01:11:20.940 --> 01:11:23.520
<v Gerald Friedland>Um, and so that's basically our work.

01:11:23.580 --> 01:11:28.800
<v Gerald Friedland>That's basically what it is, is we have, we, we basically create a solution where all this

01:11:28.800 --> 01:11:34.160
<v Gerald Friedland>configuration complexity is hidden and just basically all you need to do is do it with

01:11:34.160 --> 01:11:34.580
<v Gerald Friedland>an Arduino.

01:11:34.580 --> 01:11:38.440
<v Gerald Friedland>You know, and that is basically the, the, the, the magic here.

01:11:38.820 --> 01:11:39.080
<v Gerald Friedland>Right.

01:11:40.080 --> 01:11:40.540
<v Bertrand>Got it.

01:11:40.720 --> 01:11:40.920
<v Bertrand>Cool.

01:11:41.180 --> 01:11:41.520
<v Bertrand>Nice.

01:11:41.520 --> 01:11:45.960
<v Bertrand>Well, uh, this, so how soon is this thing coming out?

01:11:45.960 --> 01:11:50.380
<v Bertrand>Because we should probably wrap up here, but I, I, I really want to get my first thing that

01:11:50.380 --> 01:11:53.220
<v Bertrand>I want Dave to go into these so I can find out what this AGC chip is.

01:11:53.220 --> 01:11:56.860
<v Bertrand>And the other thing is I want one of these, I'm building a robotic product project right

01:11:56.860 --> 01:11:58.580
<v Bertrand>now and I want one of these for my robotics project.

01:11:58.740 --> 01:12:00.980
<v Bertrand>So how long until we can get one of these things?

01:12:01.720 --> 01:12:07.300
<v Chris Gammell>So, um, right now we're, uh, our early backers will get boards in December, the production boards

01:12:07.300 --> 01:12:07.700
<v Chris Gammell>in December.

01:12:07.940 --> 01:12:11.780
<v Chris Gammell>We will get our own, we'll get our own prototypes probably in September.

01:12:11.780 --> 01:12:16.100
<v Chris Gammell>Uh, then we get a, you know, call, you know, call and do all kinds of great thing to make

01:12:16.100 --> 01:12:17.820
<v Chris Gammell>sure it's all, uh, all good.

01:12:17.820 --> 01:12:23.500
<v Chris Gammell>But the, uh, the first 120 backers will get their board in December and then the rest will

01:12:23.500 --> 01:12:27.640
<v Chris Gammell>be available, uh, probably in February, uh, 2016.

01:12:28.620 --> 01:12:29.060
<v Bertrand>Nice.

01:12:29.300 --> 01:12:32.800
<v Chris Gammell>But, uh, the, the, the engineering, I mean, the, the hardware design is, is done at this

01:12:32.800 --> 01:12:33.080
<v Chris Gammell>point.

01:12:33.080 --> 01:12:34.440
<v Chris Gammell>Uh, you know, we're, we're done with it.

01:12:34.480 --> 01:12:40.100
<v Chris Gammell>We're, we're still, we're still doing the, the, um, you know, some of the Arduino sketch

01:12:40.100 --> 01:12:42.580
<v Chris Gammell>basically is being, uh, is being programmed at this point.

01:12:42.680 --> 01:12:44.460
<v Chris Gammell>The low level stuff is already in there.

01:12:45.020 --> 01:12:47.800
<v Chris Gammell>Uh, so, you know, as far as risk is concerned, we feel that it's, uh, it's,

01:12:47.820 --> 01:12:48.920
<v Chris Gammell>it's very much into control.

01:12:48.920 --> 01:12:52.660
<v Chris Gammell>Uh, very, uh, I mean, it's, it's, you know, we're on target basically.

01:12:53.040 --> 01:12:53.440
<v Bertrand>Yeah.

01:12:53.660 --> 01:12:57.980
<v Bertrand>And as of this recording, uh, it's 45 left of one 24.

01:12:58.500 --> 01:13:02.220
<v Bertrand>So 45 lucky people can still get that December board.

01:13:02.500 --> 01:13:04.260
<v Bertrand>And I might be, I might be one of them.

01:13:04.260 --> 01:13:05.180
<v Bertrand>There we go.

01:13:08.140 --> 01:13:09.260
<v Bertrand>Well, this is exciting.

01:13:09.440 --> 01:13:10.600
<v Dave Jones>Thank you very much for, yeah.

01:13:10.800 --> 01:13:12.380
<v Dave Jones>Thank you very much for joining us guys.

01:13:12.480 --> 01:13:14.060
<v Dave Jones>I hope your project goes well.

01:13:14.140 --> 01:13:15.400
<v Dave Jones>We'll definitely link it in.

01:13:15.680 --> 01:13:15.880
<v Dave Jones>Great.

01:13:15.980 --> 01:13:17.600
<v Dave Jones>Where can people follow you?

01:13:17.820 --> 01:13:18.700
<v Dave Jones>So on Facebook.

01:13:18.700 --> 01:13:20.300
<v Dave Jones>Are you regularly tweeting all this?

01:13:20.300 --> 01:13:20.500
<v Dave Jones>Yes.

01:13:20.640 --> 01:13:24.040
<v Dave Jones>So we have a fabulous fountain of knowledge to do with voice recognition.

01:13:24.400 --> 01:13:27.820
<v Gerald Friedland>So I, we have a Facebook page where we kind of update.

01:13:27.820 --> 01:13:28.080
<v Dave Jones>Oh, boo.

01:13:28.200 --> 01:13:29.720
<v Dave Jones>We hate Facebook here on the air, Bill.

01:13:31.160 --> 01:13:31.560
<v Gerald Friedland>Yeah.

01:13:31.620 --> 01:13:33.020
<v Gerald Friedland>Well, you know, you have to use something.

01:13:33.280 --> 01:13:35.260
<v Gerald Friedland>Um, I can tell you a lot of things about Twitter.

01:13:35.260 --> 01:13:41.220
<v Gerald Friedland>Um, um, you know, um, you, you don't, this is another, this is another podcast.

01:13:41.220 --> 01:13:45.620
<v Gerald Friedland>If I start talking about Twitter, um, I, I have a whole project on, it's called teaching

01:13:45.620 --> 01:13:46.620
<v Gerald Friedland>privacy.org.

01:13:46.620 --> 01:13:47.920
<v Gerald Friedland>So we'll talk about that later.

01:13:48.440 --> 01:13:51.180
<v Bertrand>Um, Facebook is, is stellar on their privacy stuff.

01:13:51.180 --> 01:13:51.660
<v Bertrand>Oh yeah.

01:13:51.660 --> 01:13:51.780
<v Bertrand>Yeah.

01:13:51.780 --> 01:13:51.980
<v Bertrand>Yeah.

01:13:51.980 --> 01:13:56.320
<v Gerald Friedland>Um, I'm not commenting on that either.

01:13:56.440 --> 01:13:56.780
<v Gerald Friedland>I agree.

01:13:57.000 --> 01:13:57.060
<v Gerald Friedland>Okay.

01:13:57.060 --> 01:14:02.020
<v Gerald Friedland>But the point, the point is, the point is, uh, we, you know, I have to choose some, something

01:14:02.020 --> 01:14:03.240
<v Gerald Friedland>and we chose Facebook.

01:14:03.720 --> 01:14:08.480
<v Gerald Friedland>Um, and so we have a Facebook page where we basically, this is the first source of information

01:14:08.480 --> 01:14:10.980
<v Gerald Friedland>because we put in a lot of personal stuff on there too.

01:14:10.980 --> 01:14:15.260
<v Gerald Friedland>Like, you know, we, when we did the Kickstarter video, we showed some photos of the prep for

01:14:15.260 --> 01:14:16.120
<v Gerald Friedland>this and so on.

01:14:16.120 --> 01:14:20.900
<v Gerald Friedland>So people see the early stuff, like as it's developing on that Facebook page.

01:14:21.100 --> 01:14:27.660
<v Chris Gammell>And the link is, yes, our shield slash ASR shield all in one word.

01:14:27.660 --> 01:14:27.980
<v Chris Gammell>Yeah.

01:14:28.640 --> 01:14:30.320
<v Chris Gammell>Uh, which wasn't the most obvious name.

01:14:30.420 --> 01:14:31.920
<v Chris Gammell>So let's actually go.

01:14:32.080 --> 01:14:34.320
<v Dave Jones>You can also find, so I had trouble finding that.

01:14:34.520 --> 01:14:34.620
<v Dave Jones>Yeah.

01:14:34.620 --> 01:14:39.860
<v Gerald Friedland>So ASR shield is the project name because when we started, we didn't have a fancy movie.

01:14:39.860 --> 01:14:45.440
<v Gerald Friedland>Uh, we had an ASR shield, which ASR stands for automatic speech recognition and then shield.

01:14:45.980 --> 01:14:46.380
<v Gerald Friedland>Right.

01:14:46.700 --> 01:14:49.740
<v Gerald Friedland>So ASR is actually the scientific term for the whole, all of this.

01:14:49.780 --> 01:14:51.020
<v Gerald Friedland>This is automatic speech recognition.

01:14:51.360 --> 01:14:56.320
<v Gerald Friedland>So ASR shield is what it was, you know, before we came up with that name movie.

01:14:56.580 --> 01:14:58.280
<v Chris Gammell>But you can find us on Kickstarter.

01:14:58.640 --> 01:15:06.040
<v Chris Gammell>Uh, if you search for Arduino movie, M O V I, uh, you know, maybe you can put speech in

01:15:06.040 --> 01:15:06.280
<v Chris Gammell>there.

01:15:06.280 --> 01:15:09.560
<v Bertrand>And we will of course have tons of links for the users.

01:15:09.860 --> 01:15:10.340
<v Bertrand>Yep.

01:15:10.460 --> 01:15:10.800
<v Bertrand>Thank you.

01:15:11.700 --> 01:15:12.100
<v Bertrand>Okay.

01:15:12.200 --> 01:15:12.540
<v Dave Jones>Awesome.

01:15:12.880 --> 01:15:13.940
<v Dave Jones>Thank you very much, guys.

01:15:14.060 --> 01:15:17.260
<v Bertrand>I'm very, I'm looking forward to this for robotics and everything else.

01:15:17.260 --> 01:15:23.240
<v Bertrand>I mean, I mean, uh, I, I've always wanted, uh, uh, what's the guy's name in Iron Man?

01:15:23.700 --> 01:15:24.820
<v Bertrand>The, the house thing.

01:15:26.200 --> 01:15:28.700
<v Dave Jones>Oh, um, uh, uh, uh, J.

01:15:28.900 --> 01:15:29.400
<v Dave Jones>Is that a question?

01:15:29.600 --> 01:15:29.920
<v Dave Jones>Jarvis.

01:15:30.240 --> 01:15:31.320
<v Bertrand>I want a Jarvis in my house.

01:15:31.860 --> 01:15:32.220
<v Bertrand>Jarvis.

01:15:32.220 --> 01:15:32.360
<v Bertrand>Yeah.

01:15:32.660 --> 01:15:35.980
<v Bertrand>So if this can give me a Jarvis or even something close, I'm happy.

01:15:35.980 --> 01:15:40.760
<v Gerald Friedland>So, so by the way, just on that, I tried that call sign and it worked pretty well.

01:15:41.540 --> 01:15:41.900
<v Bertrand>Okay.

01:15:42.500 --> 01:15:43.020
<v Bertrand>All right.

01:15:43.100 --> 01:15:43.380
<v Bertrand>Good.

01:15:43.540 --> 01:15:43.820
<v Bertrand>Good.

01:15:43.940 --> 01:15:44.140
<v Bertrand>Great.

01:15:44.320 --> 01:15:44.720
<v Gerald Friedland>There you go.

01:15:45.040 --> 01:15:45.840
<v Bertrand>Jarvis for the home.

01:15:47.420 --> 01:15:47.980
<v Gerald Friedland>All right.

01:15:48.100 --> 01:15:48.660
<v Bertrand>Well, thanks guys.

01:15:48.660 --> 01:15:51.640
<v Bertrand>Thanks for telling us all about this, uh, speech stuff and we'll, we'll definitely be

01:15:51.640 --> 01:15:52.100
<v Bertrand>checking it out.

01:15:52.460 --> 01:15:52.640
<v Chris Gammell>Yeah.

01:15:52.640 --> 01:15:53.100
<v Chris Gammell>Thank you.

01:15:53.300 --> 01:15:54.140
<v Chris Gammell>Thank you very much.

01:15:54.500 --> 01:15:54.740
<v Chris Gammell>See ya.

01:15:54.740 --> 01:15:54.980
<v Chris Gammell>Bye.

01:15:54.980 --> 01:15:55.460
<v Chris Gammell>Bye.

01:15:55.460 --> 01:15:55.660
<v Chris Gammell>Bye.

01:15:55.920 --> 01:15:56.360
<v Bertrand>Bye.

01:15:56.360 --> 01:15:56.440
<v Bertrand>Bye.

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:56.440
<v Speaker ?>

01:15:56.440 --> 01:15:58.440
<v Bertrand>Bye.

01:15:58.440 --> 01:15:58.440
<v Speaker ?>

01:15:58.440 --> 01:15:58.440
<v Speaker ?>

01:15:58.440 --> 01:15:58.440
<v Speaker ?>

01:15:58.440 --> 01:16:00.440
<v Speaker ?>Bye.

01:16:00.440 --> 01:16:00.440
<v Speaker ?>

01:16:00.440 --> 01:16:00.440
<v Speaker ?>

01:16:00.440 --> 01:16:02.440
<v Speaker ?>Bye.

01:16:02.440 --> 01:16:02.440
<v Speaker ?>

01:16:02.440 --> 01:16:11.360
<v Speaker ?>Bye.

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:11.360
<v Speaker ?>

01:16:11.360 --> 01:16:15.920
<v Speaker ?>x x x

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>

01:16:15.920 --> 01:16:15.920
<v Speaker ?>
