1
00:00:00,005 --> 00:00:05,525
Module 10.3, examples of start-ups that 
use signal processing as a core

2
00:00:05,525 --> 00:00:09,502
technology. 
Earlier on in this class somebody asked

3
00:00:09,502 --> 00:00:13,012
in the forum, if I follow digital signal 
processing class, can I get a job in the

4
00:00:13,012 --> 00:00:18,276
start-up and what sort of start-ups? 
So this brought us to think, well maybe

5
00:00:18,276 --> 00:00:23,550
we can describe a few start-ups that came 
out of research done in the lab.

6
00:00:23,550 --> 00:00:28,230
And our four that we discuss here, there 
are actually more that are active, but

7
00:00:28,230 --> 00:00:32,406
four will be discussed here are 
Illusonic, Quividi, Sensorscope and

8
00:00:32,406 --> 00:00:36,615
Vidinoti. 
So the first start up I want to discuss

9
00:00:36,615 --> 00:00:40,435
is called Illusonic. 
It was started by Cristof Faller who did

10
00:00:40,435 --> 00:00:43,815
his PhD thesis on a time as a 
[INAUDIBLE], and was interested in

11
00:00:43,815 --> 00:00:51,060
acoustical signal processing. 
And in particular in multi channel audio.

12
00:00:51,060 --> 00:00:54,024
So if you do acoustical signal 
processing, there are questions like

13
00:00:54,024 --> 00:00:57,670
beamforming, echo control, we just 
discussed this earlier.

14
00:00:57,670 --> 00:00:59,458
We used a project of can you hear, the 
shape of a room.

15
00:00:59,458 --> 00:01:03,658
When you want to do spatial audio 
processing, you want to generate audio

16
00:01:03,658 --> 00:01:08,942
for many channels. 
Either for headphones or for multichannel

17
00:01:08,942 --> 00:01:14,024
loudspeaker systems, you may want to do 
upmix or you take a stereo signal and you

18
00:01:14,024 --> 00:01:20,440
would like to render it as a 5:1 signal 
or as a 17:1 signal.

19
00:01:20,440 --> 00:01:25,840
And, there are tools, of course, where, 
you can use signal processing techniques,

20
00:01:25,840 --> 00:01:33,070
for example, to de-noise Music or 
de-reverb, recording of, person singing.

21
00:01:34,110 --> 00:01:38,335
So the tools that are used at Illusonic 
are classic digital signal processing

22
00:01:38,335 --> 00:01:42,820
tools, plus what was discussed briefly, 
when I talk about the class on audio and

23
00:01:42,820 --> 00:01:49,142
acoustic signal processing. 
Again, perceptual models are extremely

24
00:01:49,142 --> 00:01:52,854
important because the human auditory 
system is a very sophisticated signal

25
00:01:52,854 --> 00:01:56,856
processing device, and if you try to fool 
that device you better need to understand

26
00:01:56,856 --> 00:02:04,324
how it works. 
Here is an example of cool application,

27
00:02:04,324 --> 00:02:09,380
so let's say you have your home cinema 
and you have a stereo recording that you

28
00:02:09,380 --> 00:02:16,410
would like to listen to. 
So the home cinema has actually in this

29
00:02:16,410 --> 00:02:21,435
case one, two, three, four, five, six, 
seven, eight, nine plus probably two base

30
00:02:21,435 --> 00:02:26,310
booster somewhere, so it's probably an 
eleven channel system, so you would do

31
00:02:26,310 --> 00:02:34,460
enough mix from a stereo signal, let's 
say from your MP3 player...

32
00:02:34,460 --> 00:02:38,051
To this eleven channel spatial audio 
system, and you would like to make it so

33
00:02:38,051 --> 00:02:42,000
that it sounds really like you're in the 
concert hall.

34
00:02:42,000 --> 00:02:44,912
And so even sony has a very cool 
technology to do this, and not only do

35
00:02:44,912 --> 00:02:47,928
they have the technology, they actually 
sell a box that will do this at

36
00:02:47,928 --> 00:02:55,182
professional quality level. 
So the company is it's a small company

37
00:02:55,182 --> 00:02:59,403
about five people, half a dozen people, 
it licenses technology, state of the art

38
00:02:59,403 --> 00:03:05,775
stuff, to other, companies, and it has 
custom technologies that it develops.

39
00:03:05,775 --> 00:03:09,935
For specific applications and as I 
mentioned it has this very cool Immersive

40
00:03:09,935 --> 00:03:14,160
Audio Processor that was just launched 
this year and please visit our website

41
00:03:14,160 --> 00:03:18,060
and see this cool stuff and if you want 
to buy one of these Immersive Audio

42
00:03:18,060 --> 00:03:24,660
Processors, I can tell you it sounds 
incredibly beautiful.

43
00:03:27,010 --> 00:03:29,050
The next company I want to describe is 
Quividi.

44
00:03:29,050 --> 00:03:32,634
Now this is a very important company in 
its class because its a company of Palo

45
00:03:32,634 --> 00:03:36,794
Prandoni. 
So when he's not teaching on Coursera and

46
00:03:36,794 --> 00:03:41,530
playing his his guitar To explain signal 
processing.

47
00:03:41,530 --> 00:03:46,173
He's actually the CTO of a company in 
Paris, called Quividi.

48
00:03:46,173 --> 00:03:49,935
And Quividi does a full length thing in 
environments where you have cameras and

49
00:03:49,935 --> 00:03:54,017
you have digital signage. 
So we have advertisements on screens or

50
00:03:54,017 --> 00:03:57,702
you have information on screens, then 
Quividi clearly allows you to monitor who

51
00:03:57,702 --> 00:04:01,160
is actually watching what you are 
showing.

52
00:04:01,160 --> 00:04:04,688
So if you have a bunch of people in front 
of this camera, it will identify also

53
00:04:04,688 --> 00:04:08,216
people it will say oh, here is a lady, 
here's ladies, you also got, a few of the

54
00:04:08,216 --> 00:04:12,908
people are guys. 
It will make some statistics, how long

55
00:04:12,908 --> 00:04:16,604
the people actually watch for example in 
advertisement, where they look on the

56
00:04:16,604 --> 00:04:21,547
screen and so on. 
And this entire system is distributed in

57
00:04:21,547 --> 00:04:26,335
the cloud, and Allows you to do a 
dashboard, a so-called dashboard, of how

58
00:04:26,335 --> 00:04:31,579
your advertisement is being seen in these 
public screens, or in the malls where the

59
00:04:31,579 --> 00:04:39,392
screens are being shown. 
And at latest, they have 150 networks of

60
00:04:39,392 --> 00:04:45,334
measurements that are deployed all across 
the world as you can see.

61
00:04:45,334 --> 00:04:50,570
And there are some very famous names that 
show up and so they essentially can do

62
00:04:50,570 --> 00:04:55,113
monitoring of the quality of 
advertisement for all of these companies

63
00:04:55,113 --> 00:04:59,964
essentially in real time and provide 
reports to the effectiveness of using

64
00:04:59,964 --> 00:05:08,680
advertising on screens in public spaces. 
Okay, that's the story of Quividi.

65
00:05:08,680 --> 00:05:11,582
It's cool technology. 
It uses computer vision, image

66
00:05:11,582 --> 00:05:15,222
processing, the, it also uses a lot of, 
you know, state of the art, algorithmic

67
00:05:15,222 --> 00:05:20,344
and machine learning technology. 
Please visit their website if you want to

68
00:05:20,344 --> 00:05:26,200
know more about this one. 
The third company is called Sensorscope.

69
00:05:26,200 --> 00:05:30,808
It grew out of all the efforts of doing 
environmental monitoring and various

70
00:05:30,808 --> 00:05:34,929
projects here at DPFL. 
So, if you want to do environmental

71
00:05:34,929 --> 00:05:38,346
monitoring, it's cool if you can do real 
time visualization of what is happening

72
00:05:38,346 --> 00:05:42,204
there. 
an application where people are very

73
00:05:42,204 --> 00:05:46,551
interesting is so-called precision 
agriculture, so we want to control the

74
00:05:46,551 --> 00:05:51,230
quality, let's say for example, of water 
systems.

75
00:05:51,230 --> 00:05:55,991
you also want to detect you know, certain 
weather patterns and so on and then

76
00:05:55,991 --> 00:06:00,729
optimize crop production thanks to this 
monitoring.

77
00:06:01,750 --> 00:06:06,175
So the company does large scale sensor 
networks, deployments and data

78
00:06:06,175 --> 00:06:10,370
management. 
So you need wireless sensor networks.

79
00:06:10,370 --> 00:06:14,470
So these are small stations that talk to 
each other in an ad hoc fashion.

80
00:06:14,470 --> 00:06:17,760
So self organize sensor networks. 
Then you need signal and image

81
00:06:17,760 --> 00:06:20,314
processing. 
The usual stuff that you have learned

82
00:06:20,314 --> 00:06:22,832
here in the class. 
And of course radio communication

83
00:06:22,832 --> 00:06:27,595
technology. 
So here would be a typical example.

84
00:06:27,595 --> 00:06:32,282
you build a monitoring station. 
We have seen such monitoring stations at

85
00:06:32,282 --> 00:06:37,010
class when we have discussed sampling 
issues with respect to rain monitoring.

86
00:06:37,010 --> 00:06:42,210
So you take state-of-the-art, 
off-the-shelf sophisticated monitoring

87
00:06:42,210 --> 00:06:46,260
communication and so on. 
You build sensor stations.

88
00:06:46,260 --> 00:06:49,710
You deploy them in a self-organized 
network.

89
00:06:49,710 --> 00:06:52,870
Then from a bay station you talk to the 
cloud.

90
00:06:52,870 --> 00:06:56,814
On the cloud, you have all this data, and 
people that are interested in monitoring

91
00:06:56,814 --> 00:07:00,410
this sort of deployment get access, 
privileged access to this data and can

92
00:07:00,410 --> 00:07:04,006
take statistics and, you know, decide 
what to do, for example, about their

93
00:07:04,006 --> 00:07:12,060
precision agriculture project. 
It's a small company, half a dozen

94
00:07:12,060 --> 00:07:16,200
people, and probably its main market is 
precision all, agriculture, even if it

95
00:07:16,200 --> 00:07:19,920
started, also from a, academic point of 
view, mostly about environmental

96
00:07:19,920 --> 00:07:25,298
monitoring. 
And you can watch their website here, you

97
00:07:25,298 --> 00:07:29,453
can also watch all the data that is 
online at climaps.com.

98
00:07:29,453 --> 00:07:33,297
So all the deployments that have ever 
been done by the company and by the lab

99
00:07:33,297 --> 00:07:37,513
are actually available here on, on this 
website, and you can also use this data

100
00:07:37,513 --> 00:07:41,605
and you know, do some further signal 
processing if you're actually interested

101
00:07:41,605 --> 00:07:49,652
by this topic. 
The fourth company here is called

102
00:07:49,652 --> 00:07:53,518
Vidinoti. 
It's a recent start up from the lab and

103
00:07:53,518 --> 00:07:56,998
it works in augmented reality, in 
particle augmented reality on mobile

104
00:07:56,998 --> 00:08:00,693
devices. 
So, the core technologies image

105
00:08:00,693 --> 00:08:05,717
recognition, computer vision. 
But in a ways that is robust and then to

106
00:08:05,717 --> 00:08:10,681
also do all these processing in real time 
on small devices, like mobile phones and

107
00:08:10,681 --> 00:08:15,061
decide how much processing you do on a 
mobile phone, how much you do in the

108
00:08:15,061 --> 00:08:23,188
Cloud or on the server. 
And Vidinoti has a bunch of state of the

109
00:08:23,188 --> 00:08:27,270
art. 
Algorithms on the one side, to do

110
00:08:27,270 --> 00:08:31,716
tracking and recognition, and also a 
number of cool ideas on how to do

111
00:08:31,716 --> 00:08:37,180
augmented reality based on these 
methodologies.

112
00:08:40,510 --> 00:08:44,330
So, the technology is essentially 
cloud-based.

113
00:08:44,330 --> 00:08:48,478
But there is an iPhone application that 
you can download, and then you can

114
00:08:48,478 --> 00:08:52,966
annotate your favorite pictures or 
newspapers or whatever with augmented

115
00:08:52,966 --> 00:08:57,590
reality, and this is actually being used 
in particular in the newspaper industry

116
00:08:57,590 --> 00:09:01,806
to sort of bring digital content in a, 
you know, in a funny or attractive way

117
00:09:01,806 --> 00:09:08,815
onto a medium. 
Mainly the newspaper that is being

118
00:09:08,815 --> 00:09:12,280
challenged by, of course by Internet 
currently.

119
00:09:12,280 --> 00:09:17,920
It is a small company, less than 10 
people currently.

120
00:09:17,920 --> 00:09:22,130
It has you know, a strong research and 
development.

121
00:09:22,130 --> 00:09:26,486
It has also a strong intellectual 
portfolio based on patterns that has been

122
00:09:26,486 --> 00:09:30,689
generated over the year around Augmented 
Reality.

123
00:09:32,240 --> 00:09:36,022
And if you want to know more, here's a 
web site, and here is interactive

124
00:09:36,022 --> 00:09:40,300
application for the iPhone currently, it 
will be ported to Android within a couple

125
00:09:40,300 --> 00:09:47,134
of months as well. 
So, these were example of start-ups that

126
00:09:47,134 --> 00:09:50,704
used state of the art signal processing, 
image processing, computer vision, and so

127
00:09:50,704 --> 00:09:55,842
on. 
And bring it to the real world, in very

128
00:09:55,842 --> 00:10:02,178
concrete applications, from audio, to 
sensor networks, to augmented reality, to

129
00:10:02,178 --> 00:10:07,660
monitoring of audience in, advertising.