14. People were starting to apply IR techniques to music.
Audio ďŹles are treated like text.
FFT frames became words
Songs became âdocumentsâ
Thursday, October 22, 2009
16. Thereâs a problem with that.
Just because you can convert an mp3 to #s
doesnât mean you understand it.
Thursday, October 22, 2009
17. âMusic IRâ was born.
The applications are varied, but most
have nothing to do with music.
Thursday, October 22, 2009
18. Retrieving Music by Rhythmic Similarity
1.5
110 bpm
112 bpm
1
squared Euclidean distance
114 bpm
120 bpm
122 bpm 116 bpm
124 bpm
0.5 126 bpm
128 bpm
130 bpm
0
130 128 126 124 122 120 118 116 114 112 110
Tempo (bpm)
Figure 5. Euclidean Distance vs. Tempo
Thursday, October 22, 2009
19. The worst offender: âGenre IdentiďŹcationâ
Countless PhDs on this useless task.
Trying to teach a computer a marketing construct.
Thursday, October 22, 2009
20. Show of hands:
Is Bjork âelectronic, pop, jazzâ?
Thursday, October 22, 2009
21. At MIT I convinced someone to buy lots of computers
Thursday, October 22, 2009
31. Thereâs obviously so much more to music
than the audio signal
and that other stuff
is probably more important
Thursday, October 22, 2009
32. My brother makes music with sine waves
and nothing else
and gets a 9.7 on Pitchfork.
This is fascinating!
Thursday, October 22, 2009
33. My brother makes music with sine waves
and nothing else
and gets a 9.7 on Pitchfork.
This is fascinating!
Were the sine waves that good?
Thursday, October 22, 2009
50. But data is hard.
Most designers make very bad decisions
because their tools are inefďŹcient.
Thursday, October 22, 2009
51. Collaborative ďŹltering (X who did Y also did Z)
is so easy to make; but itâs also so terrible.
Thursday, October 22, 2009
52. Collaborative ďŹltering (X who did Y also did Z)
is so easy to make; but itâs also so terrible.
The SQL join is destroying music.
Thursday, October 22, 2009
57. In 2005 we modeled the worst case scenario:
In which collaborative ďŹltering was the only way
for an artist to get noticed.
The popular ones would eat the unknown ones alive.
3 sets of 3 artists each remained.
Thursday, October 22, 2009
58. Set A Set B Set C
Britney Spears Alice in Chains Chris Isaak
Backstreet Boys Korn Bob Dylan
Cristina Aguilera Faith no More Crowded House
Thursday, October 22, 2009
59. So the Echo Nest gives everyone great data.
They can decide on their own how to show it.
Thursday, October 22, 2009
60. The Echo Nest 2005
Somerville, MA USA
2 people
2 computers
Lots of ideas
1m documents
10,000 artists
100,000 songs
0 public facing sites
Thursday, October 22, 2009
61. The Echo Nest 2009
Somerville, MA USA
20 people
200 computers
Lots of products
5bn documents
1,000,000 artists
many millions of songs
0 public facing sites
Thursday, October 22, 2009
64. âKnow everything about music and listeners.â
âGive (and sell) great data to everyone.â
Thursday, October 22, 2009
65. âKnow everything about music and listeners.â
âGive (and sell) great data to everyone.â
âDo it automatically with no bias, on everything.â
Thursday, October 22, 2009
67. Artist Data Song Data Listener Data
⢠Tag Clouds ⢠Similar Songs ⢠demographics
⢠Similar Artists ⢠Tempo - age, gender, location
⢠Analytics ⢠Key ⢠psychographics
Familiarity Mode - preferences, lifestyle
⢠â˘
⢠music preference
⢠Hotttnesss ⢠Time Signature
Blogs Beats ⢠listening patterns
⢠â˘
News Downbeats ⢠tastemaker proďŹling
⢠â˘
- writers, bloggers
⢠Reviews ⢠Segments
⢠Audio ⢠Timbre
⢠Video ⢠Pitch
⢠ProďŹle Sites ⢠Loudness
⢠Misspellings ⢠Sections
⢠Aliases
Thursday, October 22, 2009
68. We have a lot of data and
we have a lot of products.
We sell mostly to social networks, labels;
video games; PR ďŹrms; musicians
Thursday, October 22, 2009
69. Artist metrics
Acoustic
Similarity analysis
Feeds
Remix
Metadata
Search / Tags
Predictive Recommendation
analytics
Thursday, October 22, 2009
70. The reason we are special is 2 things:
Scale and Platform
Thursday, October 22, 2009
71. Our scale is limitless.
We have hundreds of computers
We always do our computation on everything.
We can learn about new music very quickly.
Thursday, October 22, 2009
72. Scale
All Music Guide Pandora The Echo Nest
known artists 280,000 80,000 1,000,000
years to get there 18 8 1
time to understand
1 week 1 day <1 minute
one album
cost to understand
$400 $40 $0.001
one album
Thursday, October 22, 2009
73. Our platform is huge.
We have thousands of âfreeâ developers using our API
Our customers use the same platform
So do we.
Thursday, October 22, 2009
75. We sell two main products:
Fanalytics is a predictive analytics toolset for artists
The Knowledge is a dynamic metadata service
(recommendation, feeds, data)
for web sites
Thursday, October 22, 2009
76. Fanalytics lets artists and labels get a view
into the world of online music
We recommend blogs for artists
We show predicted analytics on activity
Thursday, October 22, 2009
77. Predictive
analytics
Artist metrics
Thursday, October 22, 2009
78. We also maintain a popular open source
remixing community and code base
so people can make awesome free
mashups, remixes, web sites using our tech
Not much of a business but we love it.
Thursday, October 22, 2009
82. Morecowbell.dj adds cowbell to any song
This Is My Jam was a pre-Muxtape (by one day)
mixtape sharing site that only let you use 30s samples
and made a total mess of the output.
Like I said, not much of a business.
Thursday, October 22, 2009
85. We also have artists using Remix
-- our data is now powering some next generation
electronic music
Thursday, October 22, 2009
86. Iâve always wanted to hear Michael
Jackson trying to sing Amerieâs âOne
Thingâ automatically by comparing
timbre, pitch and loudness distances.
-B.L.
Thursday, October 22, 2009
92. The Knowledge is a much better music data service
Customers can subscribe to constantly-updated
similarity, metadata, feeds, recommendations, etc
Thursday, October 22, 2009
93. Our similarity and recommendation data
is some of the best, because we use so many sources
and we know about all artists even if they are tiny
Thursday, October 22, 2009
95. Since our similarity is based on so many features:
popularity, audio analysis, text analysis,
structured metadata, inďŹuences, ...
Thursday, October 22, 2009
96. Since our similarity is based on so many features:
popularity, audio analysis, text analysis,
structured metadata, inďŹuences, ...
We provide our customers with the knobs
and let them decide what is important for the task.
Thursday, October 22, 2009
97. Since our similarity is based on so many features:
popularity, audio analysis, text analysis,
structured metadata, inďŹuences, ...
We provide our customers with the knobs
and let them decide what is important for the task.
We do not give a âsingle answer.â
There is no single answer.
Thursday, October 22, 2009
103. Weâve been running large scale data mining
on millions of listeners to help with analytics,
for example a gender predictor based on your music taste
Thursday, October 22, 2009
104. Hereâs the basis vectors; strongest correlators of gender:
Thursday, October 22, 2009
105. Male Female
Pet Shop Boys Eternal
Fort Minor Metro Station
Justice Gackt
Mike OldďŹeld Paolo Nutini
U2 London after Midnight
Thursday, October 22, 2009
106. 2. More musicians to use our remix tools
Thursday, October 22, 2009
107. (Iâve noticed the better you are with computers,
the worse your music is. This may just be me)
Thursday, October 22, 2009
108. 100%
75%
Music goodness
50%
25%
0%
nothing not much a little somewhat pretty good expert dork prime
Computers know-how
Thursday, October 22, 2009
110. We will soon make all of our acoustic data
available for searching and browsing
(right now it has to be your content):
âFind me a drum hit in this collection
that sounds like the break in âSingle Ladiesââ
Thursday, October 22, 2009