How AI could compose a personalized soundtrack to your life | Pierre Barreau

167,508 views ・ 2018-10-01

TED


Please double-click on the English subtitles below to play the video.

00:12
About two and a half years ago, I watched this movie called "Her."
0
12913
3428
00:16
And it features Samantha, a superintelligent form of AI
1
16365
5127
00:21
that cannot take physical form.
2
21516
1760
00:23
And because she can't appear in photographs,
3
23858
2230
00:26
Samantha decides to write a piece of music
4
26112
2016
00:28
that will capture a moment of her life just like a photograph would.
5
28152
3466
00:32
As a musician and an engineer, and someone raised in a family of artists,
6
32953
4408
00:37
I thought that this idea of musical photographs was really powerful.
7
37385
4176
00:41
And I decided to create an AI composer.
8
41955
2868
00:44
Her name is AIVA, and she's an artificial intelligence
9
44847
3488
00:48
that has learned the art of music composition
10
48359
2393
00:50
by reading over 30,000 scores of history's greatest.
11
50776
2785
00:54
So here's what one score looks like to the algorithm
12
54165
2456
00:56
in a matrix-like representation.
13
56645
2202
00:58
And here's what 30,000 scores,
14
58871
2246
01:01
written by the likes of Mozart and Beethoven,
15
61141
2096
01:03
look like in a single frame.
16
63261
2031
01:07
So, using deep neural networks, AIVA looks for patterns in the scores.
17
67609
4340
01:12
And from a couple of bars of existing music,
18
72260
3563
01:15
it actually tries to infer what notes should come next in those tracks.
19
75847
3708
01:19
And once AIVA gets good at those predictions,
20
79887
2498
01:22
it can actually build a set of mathematical rules
21
82409
3613
01:26
for that style of music
22
86046
1166
01:27
in order to create its own original compositions.
23
87236
2788
01:30
And in a way, this is kind of how we, humans, compose music, too.
24
90490
3709
01:34
It's a trial-and-error process,
25
94561
1492
01:36
during which we may not get the right notes all the time.
26
96077
3008
01:39
But we can correct ourselves,
27
99109
1389
01:40
either with our musical ear or our musical knowledge.
28
100522
3293
01:45
But for AIVA, this process is taken from years and years of learning,
29
105495
4143
01:49
decades of learning as an artist, as a musician and a composer,
30
109662
3262
01:52
down to a couple of hours.
31
112948
1733
01:55
But music is also a supersubjective art.
32
115030
2880
01:57
And we needed to teach AIVA
33
117934
1564
01:59
how to compose the right music for the right person,
34
119522
2445
02:01
because people have different preferences.
35
121991
2141
02:04
And to do that, we show to the algorithm over 30 different category labels
36
124156
4190
02:08
for each score in our database.
37
128370
1945
02:10
So those category labels are like mood
38
130339
2516
02:12
or note density or composer style of a piece
39
132879
2920
02:15
or the epoch during which it was written.
40
135823
2524
02:18
And by seeing all this data,
41
138371
1935
02:20
AIVA can actually respond to very precise requirements.
42
140330
3056
02:23
Like the ones, for example, we had for a project recently,
43
143966
3103
02:27
where we were commissioned to create a piece
44
147093
3035
02:30
that would be reminiscent of a science-fiction film soundtrack.
45
150152
3027
02:33
And the piece that was created is called "Among the Stars"
46
153839
4500
02:38
and it was recorded with CMG Orchestra in Hollywood,
47
158363
3317
02:41
under great conductor John Beal,
48
161704
1714
02:43
and this is what they recorded, made by AIVA.
49
163442
3067
02:47
(Music)
50
167657
7000
03:30
(Music ends)
51
210196
2190
03:34
What do you think?
52
214719
1207
03:35
(Applause)
53
215950
4380
03:40
Thank you.
54
220354
1150
03:42
So, as you've seen, AI can create beautiful pieces of music,
55
222001
4055
03:46
and the best part of it
56
226080
1803
03:47
is that humans can actually bring them to life.
57
227907
2581
03:51
And it's not the first time in history
58
231506
1838
03:53
that technology has augmented human creativity.
59
233368
3344
03:56
Live music was almost always used in silent films
60
236736
3072
03:59
to augment the experience.
61
239832
1666
04:01
But the problem with live music is that it didn't scale.
62
241522
3182
04:04
It's really hard to cram a full symphony into a small theater,
63
244728
3429
04:08
and it's really hard to do that for every theater in the world.
64
248181
2991
04:11
So when music recording was actually invented,
65
251196
2849
04:14
it allowed content creators, like film creators,
66
254069
2349
04:16
to have prerecorded and original music
67
256442
2786
04:19
tailored to each and every frame of their stories.
68
259252
2873
04:22
And that was really an enhancer of creativity.
69
262149
2626
04:26
Two and a half years ago, when I watched this movie "Her,"
70
266617
3286
04:29
I thought to myself that personalized music
71
269927
2907
04:32
would be the next single biggest change in how we consume and create music.
72
272858
4417
04:38
Because nowadays, we have interactive content, like video games,
73
278187
4222
04:42
that have hundreds of hours of interactive game plays,
74
282433
3015
04:45
but only two hours of music, on average.
75
285472
2113
04:47
And it means that the music loops and loops and loops
76
287609
2492
04:50
over and over again, and it's not very immersive.
77
290125
2332
04:52
So what we're working on is to make sure that AI can compose
78
292467
3951
04:56
hundreds of hours of personalized music
79
296442
2302
04:58
for those use cases where human creativity doesn't scale.
80
298768
3866
05:03
And we don't just want to do that for games.
81
303363
2206
05:06
Beethoven actually wrote a piece for his beloved, called "Für Elise,"
82
306657
4428
05:11
and imagine if we could bring back Beethoven to life.
83
311109
3651
05:14
And if he was sitting next to you, composing a music for your personality
84
314784
5230
05:20
and your life story.
85
320038
1333
05:22
Or imagine if someone like Martin Luther King, for example,
86
322632
2779
05:25
had a personalized AI composer.
87
325435
2039
05:27
Maybe then we would remember
88
327498
1373
05:28
"I Have a Dream" not only as a great speech,
89
328895
2056
05:30
but also as a great piece of music, part of our history,
90
330975
2626
05:33
and capturing Dr. King's ideals.
91
333625
2133
05:36
And this is our vision at AIVA:
92
336069
1889
05:37
to personalize music so that each and every one of you
93
337982
2578
05:40
and every individual in the world
94
340584
1649
05:42
can have access to a personalized live soundtrack,
95
342257
3074
05:45
based on their story and their personality.
96
345355
2505
05:49
So this moment here together at TED is now part of our life story.
97
349915
4079
05:54
So it only felt fitting that AIVA would compose music for this moment.
98
354018
4169
05:58
And that's exactly what we did.
99
358674
2444
06:01
So my team and I worked on biasing AIVA on the style of the TED jingle,
100
361436
5022
06:06
and on music that makes us feel a sense of awe and wonder.
101
366482
3444
06:09
And the result is called "The Age of Amazement."
102
369950
3913
06:13
Didn't take an AI to figure that one out.
103
373887
2396
06:16
(Laughter)
104
376307
1150
06:18
And I couldn't be more proud to show it to you,
105
378152
2402
06:20
so if you can, close your eyes and enjoy the music.
106
380578
2428
06:23
Thank you very much.
107
383030
1333
06:25
(Music)
108
385537
4195
06:35
[The Age of Amazement Composed by AIVA]
109
395176
2667
08:19
(Music ends)
110
499649
1151
08:20
This was for all of you.
111
500824
1373
08:22
Thank you.
112
502221
1166
08:23
(Applause)
113
503411
4573
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7