Method and system for music recommendation based on immunology

1. A method of recommending music comprising:

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

b) mathematically analyzing each digital song file to determine a numerical value for each of a plurality of selected quantifiable characteristics, wherein each selected quantifiable characteristic is a physical parameter based on human perception;

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

d) extracting a plurality of music descriptors for each song vector;

e) selecting a seed song and comparing the song vector for said seed song to the song vector for each said digital song file by summing the square of the difference between the numerical values of each characteristic in each said song vector based on the music descriptors in step d); and

f) deriving a list of songs wherein the sum of the square of the difference between the numerical value of each characteristic based on the music descriptors in each said song vector is below a predetermined threshold.

2. The method according to claim 1 , said comparing song vectors comprising evaluating a Euclidean distance between said song vectors.

3. The method according to claim 1 , wherein the step of extracting a plurality of music descriptors for each song vector uses Principal Component Analysis in order to minimize the plurality of descriptors vector space cardinality, eliminating correlated dimensions.

4. The method according to claim 1 , further comprising the steps of:

g1) creating, from the list of songs determined in step f), pairs of similar and non-similar songs, wherein similar comprises a small distance according to a Riemann Metric in a curved space of sonic descriptors;

h1) using an iterative method, which learns the Riemannian Metric, determining pairs of songs having sonic descriptors most similar to the seed song;

i1) maintaining the pairs of songs in a training database; and

j1) deriving a list of songs having the least distance in the Riemannian metric in pairs of songs from the seed song.

5. The method according to claim 1 , further comprising the steps of:

g) transcoding each song in the digital database to discard nonsignificant audio data;

h) dividing each transcoded song into a plurality of millisounds;

i) characterizing each millisound by a unique audio-fingerprint;

j) establishing a plurality of song supergenres having similar audio-fingerprints; and

k) classifying each supergenre wherein the millisounds include most of the music descriptors.

6. The method according to claim 5 , further comprising the steps of:

l) comparing the millisounds from the seed song to selected supergenres wherein the millisounds in the selected supergene include most of the music descriptors to determine a probability of inclusion of the seed song in said selected supergenre, wherein said comparison comprises a clustering algorithm employing the Riemann Metric where the supergenres minimize intradistance and maximize interdistance;

m) comparing the millisounds from each song in the list created in step f) to the selected supergenres wherein the millisounds in the selected supergene include most of the music descriptors to determine which songs are most different from said selected supergenre according to the Riemann Metric, maintaining songs having sonic descriptors most similar to the seed song in a training database; and

n) designating those songs that are most different from said selected supergenre as immune songs.

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

o) comparing the millisounds from the seed song to the millisounds for the songs in the training database;

p) discarding from the training database those songs having similar millisounds, wherein similar means a small distance in the Riemann Metric;

q) comparing the millisounds for songs remaining in the training database with the millisounds for the immune songs; and

r) discarding from the list generated in step f) those songs having most millisounds in common with the immune songs.

8. The method according to claim 5 , said step of dividing each song into a plurality of millisounds further comprising detecting changes in music patterns by isolating changes in characteristic parameters of the sound signal.

9. A computer implemented method of determining a user's preference of music, comprising the steps of:

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

providing an analysis engine having software for use in a computer processor adapted to execute said software;

using said computer processor to analyze each digital song file to determine a numerical value for each of a plurality of quantifiable characteristics, wherein each quantifiable characteristic is a physical parameter based on human perception;

using said computer processor to create a multidimensional song vector for each digital song file, said multidimensional song vector representing numerical values for each of said quantifiable characteristics;

using said computer processor to extract a plurality of music descriptors for each song vector;

selecting a seed song and using said computer processor to compare the song vector for said seed song to the song vector for each song file by summing the square of the difference between the numerical values of each characteristic in each said song vector based on the music descriptors; and

using said computer processor to derive a first list of songs wherein the sum of the square of the difference between the numerical value of each characteristic based on the music descriptors in each said song vector is below a predetermined threshold.

10. The method according to claim 9 , said using said computer processor to compare song vectors further comprising evaluating a Euclidean distance between said song vectors.

11. The method according to claim 9 , wherein the step of using said computer processor extracting a plurality of music descriptors for each song vector further comprises said computer processor using Principal Component Analysis in order to minimize the plurality of descriptors vector space cardinality, eliminating correlated dimensions.

12. The method according to claim 9 , further comprising the steps of:

said computer processor creating, from the list of songs, pairs of similar and non-similar songs, wherein similar comprises a small distance according to a Riemann Metric in a curved space of sonic descriptors;

said computer processor, using an iterative method which learns the Riemannian Metric, determining the pairs of songs having sonic descriptors most similar to the seed song;

storing an identification for the pairs of songs in a training database; and

using said computer processor to derive a list of songs having the least distance in the Riemannian metric in pairs of songs from the seed song.

13. The method according to claim 9 , further comprising the steps of:

using said computer processor to transcode each song in the digital database in order to discard nonsignificant audio data;

said computer processor dividing each transcoded song into a plurality of millisounds;

using said computer processor to establish a plurality of song supergenres having similar audio-fingerprints; and

classifying each supergenre wherein the millisounds include most of the music descriptors.

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

said computer processor comparing the millisounds from the seed song to selected supergenres wherein the millisounds in the selected supergene include most of the music descriptors to determine a probability of inclusion of the seed song in said selected supergenre, wherein said comparison comprises a clustering algorithm employing the Riemann Metric where the supergenres minimize intradistance and maximize interdistance, maintaining songs having sonic descriptors most similar to the seed song in a training database;

said computer processor comparing the millisounds from each song in the first list of songs to the selected supergenres wherein the millisounds in the selected supergene include most of the music descriptors to determine which songs are most different from said selected supergenre according to the Riemann Metric; and

designating those songs that are most different from said selected supergenre as immune songs.

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

said computer processor comparing the millisounds from the seed song to the millisounds for the songs in the training database;

said computer processor discarding from the training database those songs having similar millisounds, wherein similar means a small distance in the Riemann Metric;

said computer processor comparing the millisounds for songs remaining in the training database with the millisounds for the immune songs; and

discarding from the first list of songs those songs having most millisounds in common with the immune songs.

16. The method according to claim 13 , said step of dividing each song into a plurality of millisounds further comprising detecting changes in music patterns by isolating changes in characteristic parameters of the sound signal.