Method and system for music recommendation

• US 7,081,579 B2
• Filed: 10/03/2003
• Issued: 07/25/2006
• Est. Priority Date: 10/03/2002
• Status: Active Grant

• Current Assignee: Music Intelligence Solutions, Inc. (Akazoo S.A.)
• Original Assignee: Polyphonic Human Media Interface SL (Polyphonic Human Media Interface)
• Inventors: Llopis, Antonio Trias, Ysern, Daniel Caldentey, Ullod, Carlos María López, Arkwright, Dominic, Marcos, Jesús Sanz, Bonet, Antonio Trias, Alcalde, Vicenç Gaitan
• Primary Examiner(s): Fletcher, Marlon T.
• Assistant Examiner(s): Qin, Jianchun
• Application Number:
  US10/678,505
• Publication Number:
  US 20040107821A1
• Time in Patent Office: 1,026 Days
• Field of Search: 84604-609, 846/23, 846/26, 846/60, 846/61, 846/93, 84697-699, 708/403, 707/1, 700/94, 705/10, 713/176, 434/307.A
• US Class Current: 84/608
• CPC Class Codes: G06F 16/634 Query by example, e.g. query by humming
• CPC Class Codes: G06F 16/637 Administration of user profiles, e.g. generation, initialization, adaptation or distribution
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content
• CPC Class Codes: G06F 16/686 using information manually generated, e.g. tags, keywords, comments, title or artist information, time, location or usage information, user ratings
• CPC Class Codes: G10H 1/0058 Transmission between separate instruments or between individual components of a musical system G10H1/0083 takes precedence
• CPC Class Codes: G10H 2210/036 of musical genre, i.e. analysing the style of musical pieces, usually for selection, filtering or classification
• CPC Class Codes: G10H 2240/105 User profile, i.e. data about the user, e.g. for user settings or user preferences
• CPC Class Codes: G10H 2250/235 Fourier transform; Discrete Fourier Transform [DFT]; Fast Fourier Transform [FFT]

Abstract

An artificial intelligence song/music recommendation system and method is provided that allows music shoppers to discover new music. The system and method accomplish these tasks by analyzing a database of music in order to identify key similarities between different pieces of music, and then recommends pieces of music to a user depending upon their music preferences. An embodiment enables a user to evaluate a new song'"'"'s similarity to songs already established as commercially valuable.

Claims

1. A method of analyzing music, said method comprising the steps of:

a) providing a digital database comprising a plurality of digital song files;

b) selecting one of said song files for analysis;

c) dividing said selected song file into a plurality of discrete parts;

d) using Fast Fourier Transform techniques on each part of said selected song file to establish a plurality of coefficients, wherein said coefficients are representative of predetermined quantifiable characteristics of said selected song, and;

wherein each said predetermined characteristic is a physical parameter based on human perception including;

brightness;

bandwidth;

tempo;

volume;

rhythm;

low frequency

octave, and

how said parameters change over time;

e) determining an average value of the coefficients for each characteristic from each said part of said selected song file;

f) compiling a song vector comprising a sequential list of said average values of the coefficients for each said characteristic for said selected song file; and

g) repeating steps b) through f) for each song in said database.

2. The method according to claim 1, wherein said digital database comprises a plurality of compressed digital song files, said method further comprising the step of:

b1) decompressing said selected song file prior to dividing said selected song file into a plurality of discrete parts.

3. The method according to claim 1, wherein said digital song files are formatted as linear PCM audio data.

4. A method of determining a user'"'"'s music preference, said method comprising the steps of:

a) providing a digital database comprising a plurality of digital song files;

b) mathematically analyzing each said digital song file to determine a numerical value for a plurality of selected quantifiable characteristics;

wherein each said characteristic is a physical parameter based on human perception including;

brightness;

bandwidth;

tempo;

volume;

rhythm;

low frequency;

noise; and

octave, and

how said parameters change over time;

c) compiling a song vector comprising a sequential list of said numerical values for each of said plurality of selected characteristic for each said song file;

d) dividing each said song file into portions of selected size and mathematically analyzing each said portion to determine a numerical value for said plurality of selected characteristics for each said portion and compiling a portion vector comprising a sequential list of numerical values for each of said plurality of characteristics for each said portion;

e) selecting and storing a representative portion of each said song file wherein the portion vector of said representative portion substantially mathematically matches the song vector of said song file;

f) choosing two dissimilar representative portions and enabling said user to listen to both representative portions;

g) permitting said user to indicate which of said two dissimilar representative portions said user prefers; and

h) repeating steps f) and g), as necessary, to establish a taste vector for said user comprising song characteristics that said user prefers.

5. The method according to claim 4, said mathematically analyzing steps further comprising the step of:

using fast Fourier Transform techniques to establish a plurality of coefficients, wherein said coefficients are representative of said characteristics of said song.

6. The method according to claim 4, further comprising the steps of:

i) normalizing and centering the data for each characteristic across the entire database to determine the most useful information from each characteristic;

j) comparing said user'"'"'s taste vector to each said song vector by summing the square of the difference between the numerical values of each characteristic in each said vector; and

k) recommending to said user, a list of at least one song wherein the sum of the square of the difference between the numerical value of each characteristic in each said vector is below a predetermined threshold.

7. The method according to claim 6, further comprising the steps of:

l) enabling said user to listen to a song from said list of recommended songs and permitting said user to select to listen to more songs similar to said selection;

m) comparing the song vector of said selected song to the normalized and centered song vector in the database by summing the square of the difference between the numerical values of each characteristic in each said vector; and

n) recommending to said user, at least one song wherein the sum of the square of the difference between the numerical value of each characteristic in each said vector is below a predetermined threshold.

8. The method according to claim 4, wherein said method is performed using a real-time process based on dynamic, interaction with said user.

9. The method according to claim 8, wherein upon permitting said user to indicate which of two dissimilar representative portions said user prefers, said user chooses either which portion is liked best or which portion is disliked least;

then, according to said user'"'"'s preference selection based on the portions presented to the user, selecting the next set of representative portions to present to the user in a dynamic manner; and

said step of establishing a taste vector for said user further comprising the steps of;

continuing to present sets of representative portions of songs until a learning process is completed and a profile of user taste can be established for recommendation of songs according to said profile.

10. The method according to claim 8, wherein said real-time, interactive process with said user is performed over a computer network.

11. A method of determining a user'"'"'s music preference, said method comprising the steps of:

a) providing a digital database comprising a plurality of digital song files;

b) providing to said user a list of a plurality of songs selected from said database;

c) permitting said user to subjectively label each song in said list of a plurality of songs according to said user'"'"'s likes and dislikes;

d) analyzing said song'"'"'s from said list of a plurality of songs wherein said song'"'"'s having been indicated as ‘

liked’

by said user are separately analyzed from said song'"'"'s having been indicated as ‘

disliked’

by said user, said analysis of said song'"'"'s further comprising the steps of;

d1) dividing each said song into a plurality of discrete parts;

d2) using Fast Fourier Transform techniques on each part of said song to establish a plurality of coefficients, wherein each said coefficient is representative of a quantifiable characteristic of said selected song, and;

wherein each said characteristic is a physical parameter based on human perception including;

brightness;

bandwidth;

tempo;

volume;

rhythm;

low frequency;

noise; and

octave, and

how said parameters change over time;

d3) determining an average value for each coefficient from each said part of said song;

d4) compiling a song vector for each said song comprising a sequential list of said average values for each said coefficient for said selected song file; and

d5) repeating steps d1) through d4) for each song in said list of a plurality of songs; and

e) determining a user taste profile according to songs having been indicated as ‘

liked’

by said user.

12. The method according to claim 11, wherein said subjective labels are selected from the group consisting of:

“

I strongly like it”

;

“

I somewhat like it”

;

“

I neither like nor dislike it”

;

“

I dislike it”

; and

“

I strongly dislike it”

.

13. The method according to claim 11, the step of determining a user taste profile further comprising the steps of:

compiling a user taste vector comprising a sequential list of values for each coefficient wherein each such coefficient is weighted according to said user'"'"'s subjective likes and dislikes.

14. The method according to claim 13 , further comprising the steps of:

f) normalizing and centering the data for each coefficient across the entire database to determine the most useful information from each characteristic coefficient;

g) comparing said user'"'"'s taste vector to each said song vector by summing the square of the difference between the numerical values of each coefficient in each said vector; and

h) recommending to said user, a list of at least one song wherein the sum of the square of the difference between the numerical value of each coefficient in each said vector is below a predetermined threshold.

15. The method according to claim 14, further comprising the steps of:

i) enabling said user to listen to a song from said list of recommended songs and permitting said user to select to listen to more songs similar to said selection;

j) comparing the song vector of said selected song to the song vector in the database by summing the square of the difference between the numerical values of each characteristic coefficient in each said vector; and

k) recommending to said user, at least one song wherein the sum of the square of the difference between the numerical value of each characteristic coefficient in each said vector is below a predetermined threshold.

16. The method according to claim 11, wherein the steps of determining a user taste profile is performed using a real-time process based on dynamic, interaction with said user.

17. The method according to claim 16, steps of determining a user taste profile further comprising the step of:

e1) selecting and storing a representative portion of each said song file wherein said representative portion substantially mathematically matches said song file;

e2) choosing two dissimilar representative portions and enabling said user to listen to both representative portions;

e3) permitting said user to indicate which of the two dissimilar representative portions said user prefers, wherein said user chooses either which portion is liked best or which portion is disliked least;

e4) then, according to said user'"'"'s preference selection based on the portions presented to the user, selecting another set of representative portions to present to the user in a dynamic manner; and

e5) continuing to present sets of representative portions of songs until a learning process is completed and a profile of user taste can be established for recommendation of songs according to said profile.

18. The method according to claim 16, wherein said real-time, interactive process with said user is performed over a computer network.

19. A method of comparing a new song to previously commercially successful songs, said method comprising:

a) establishing a digital database comprising a plurality of digital song files wherein said songs have been identified as commercially successful;

b) mathematically analyzing each said digital song file to determine a numerical value for a plurality of selected Quantifiable characteristics;

wherein each said characteristic is a physical parameter based on human perception including;

brightness;

bandwidth;

tempo;

volume;

rhythm;

low frequency;

noise; and

octave, and

how said parameters change over time;

c) compiling a song vector comprising a sequential list of said numerical values for each of said plurality of selected characteristic for each said song file;

d) presenting said new song as a digital music file for comparison;

e) mathematically analyzing said new song file to determine a numerical value for the same plurality of selected quantifiable characteristics;

f) compiling a new song vector comprising a sequential list of said numerical values for each of said plurality of selected characteristic for said new song file;

g) establishing an affinity value for said new song as compared to each song vector in the database by summing the square of the difference between the numerical values of each characteristic in each said vector; and

j) determining the potential for commercial success if said affinity value is below a predetermined threshold.

20. The method according to claim 19, said mathematically analyzing steps further comprising the step of:

using fast Fourier Transform techniques to establish a plurality of coefficients, wherein said coefficients are representative of said characteristics of said song.