SortMyTunes

1
SortMyTunes

• Martin McCrory
• School of Informatics, Indiana University
  Bloomington
• April 18, 2008
• This presentation and any accompanying commentary
  is released to the world under the GNU General
  Public License on April 18, 2008.

2

• SortMyTunes is a framework for metadata-based
  Music Classification.

Why is this kind of classification important?
Most music collections have at least some
metadata component!
3

• So, how does SortMyTunes work?

• The user creates pods, or bins of music that
  are different from each other in some way.
• SMT classifieds a portion of the rest of the
  database into these pods.
• The user makes manual corrections, and then SMT
  iterates again.

4

• Where does the metadata come from?

Manual metadata creation
Etc.
5

• What are tags?

• Tags form the basic building block of metadata
  for each track.
• The use of tags as metadata replaces any need to
  retrieve data from actual audio.

6

• Example

User creates a file representing the initial
state of the pods
Alternative!, artist1, title1, Guitar, 2000s,
Attractive, Acoustic Rap-tastic, artist2,
title2, Rap, Doggy, Style, Annoying Classical
and Rock, artist3, title3, Classical, Symphony,
Strings, Long, Annoying, Boring Classical and
Rock, artist4, title4, rock, pop
7

• Example

Example
User inputs a file that represents the rest of
the database
Jack Johnson, Where's My Girl, Guitar, 2000s,
Attractive, Acoustic Beyonce, Dontcha, Rap,
Doggy, Style, Annoying Beethoven, 5th String
Quartet, Classical, Symphony, Strings, Long,
Annoying, Boring John Mayer, Blibbity Blah,
guitar, long, aCousTiC Brahms, Symphony,
Classical, Symphony, Orchestra, Chicago Snoop
Dogg, Thugz, Rap, Doggy, Style, RB Indigo Girls,
Go Girls!, Pop, Live, Concert, Loud, Rock Dave
Matthews, Crash, Guitar, Acoustic, Live, Lame,
Drama, Violin Stravinsky, Rite of Spring,
Classical, Symphony, Loud, Intense Wesley Willis,
Who Gives a Shit, Electronic, Repetitive, Dumb,
Lame No Doubt, Gwen Stefani's Song, Pop, Rock,
Woman, Vocals, Attractive Aphex Twin,
Windowlicker, Ambient, Electronic, Techno,
Bizarre, Wierd Aphex Twin, Orange, Orchestral,
Orchestra, Ambient, Electronic, Techno,
Irregular, Time, Signature Aphex Twin,
Dijeridoo, Ambient, Dijeridoo, Instrument,
Instrumental, Rhythmic Chopin, Piano Etude,
Piano, Etude, Chopin, Classical, Harmony,
Gay Kanye West, I'm So Awesome, Rap, Rock, RB,
Black, Rapper
8

• Example

SMT sorts the collection into the pods that the
user specified
---Alternative!--- Pod Tracks
artist1 - title1 2000s, Attractive, Acoustic, Jack Johnson -
Where's My Girl Acoustic, John Mayer - Blibbity Blah guitar, long, aCousTiC, Dave Matthews - Crash
Violin, Wesley Willis - Who Gives a Shit

Aphex Twin - Windowlicker Electronic, Techno, Bizarre, Wierd, Aphex
Twin - Orange Ambient, Electronic, Techno, Irregular, Time,
Signature, Aphex Twin - Dijeridoo Ambient, Dijeridoo, Instrument, Instrumental,
Rhythmic, Pod Tags (with count)
(Acoustic4), (Guitar4),
(Ambient3), (Electronic3), (Techno2),
(Lame2), (Attractive2), (2000s2),
(Rhythmic1), (Instrumental1), (Instrument1),
(Dijeridoo1), (Signature1), (Time1),
(Irregular1), (Orchestra1), (Orchestral1),
(Wierd1), (Bizarre1), (Dumb1), (Repetitive1)
9

• Example

SMT sorts the collection into the pods that the
user specified
---Rap-tastic--- Pod Tracks
artist2 - title2 Doggy, Style, Annoying, Beyonce - Dontcha
Snoop
Dogg - Thugz
Kanye West - I'm So Awesome RB, Black, Rapper, Pod Tags (with
count) (Rap4), (Style3), (Doggy3),
(RB2), (Annoying2), (Rapper1), (Black1),
(Rock1)
10
Example
SMT sorts the collection into the pods that the
user specified
---Classical and Rock--- Pod
Tracks artist3 - title3 Classical, Symphony, Strings, Long, Annoying,
Boring, artist4 - title4 Beethoven - 5th String Quartet Classical, Symphony, Strings, Long, Annoying,
Boring, Brahms - Symphony Symphony, Orchestra, Chicago, Indigo Girls -
Go Girls! Stravinsky - Rite of Spring Symphony, Loud, Intense, No Doubt - Gwen
Stefani's Song Attractive, Chopin - Piano Etude Piano, Etude, Chopin, Classical, Harmony, Gay,
Pod Tags (with count)
(Classical5), (Symphony4), (pop3), (rock3),
(Loud2), (Boring2), (Annoying2), (Long2),
(Strings2), (Gay1), (Harmony1), (Chopin1),
(Etude1), (Piano1), (Attractive1), (Vocals1),
(Woman1), (Intense1), (Concert1)
11

• Miscellanea

• SMT uses the k-means classification algorithm.
• SMT is not limited with regard to collection
  size.
• Current efficiency of SMT is O(n2).

12
Questions/Contact

• Martin McCrory
• mccrory_at_indiana.edu