WO2007045797A1

WO2007045797A1 - Method, program and device for describing a music file, method and program for comparing two music files with one another, and server and terminal for carrying out these methods

Info

Publication number: WO2007045797A1
Application number: PCT/FR2006/051057
Authority: WO
Inventors: Luigi Lancieri; Lucille Tanquerel
Original assignee: France Telecom
Priority date: 2005-10-20
Filing date: 2006-10-18
Publication date: 2007-04-26

Abstract

In this describing method, a description vector is used that contains at least one description coordinated defined with the aid of at least one parameter calculated from a frequency spectrum of at least one segment (s1, s2, s3) of a sample (E) of a signal (S) representing the music file. A first calculated parameter corresponds to a directional coefficient of a regression line of points of a two-dimensional space, the coordinates of these points being frequencies and amplitudes defining the frequency spectrum.

Description

Method, program and device for describing a musical file, method and program for comparing two music files with each other, and server and terminal for the implementation of these methods.

The present invention relates to a method, a program and a device for describing a musical file, a method and a program for comparing two music files with each other, and a server and a terminal for implementing these methods.

In order to perform automatic processing of audio data, such as music file rankings by music genre in a database comprising such files or comparisons of several music files with each other, it is known to associate respective descriptors with these music files, each comprising at least one description vector.

In this case, the classification or the comparison of the musical files is carried out using these vectors of description.

Typical coordinates of such a description vector are, for example, the Centroid, the Rolloff, the ZeroCrossings or the MFCCs (Mel-Frequency Cepstral Coefficients). Some of these description coordinates are determined using a frequency spectrum of at least one sample of a signal representing the musical file. Thus, in the state of the art, a method of describing a musical file, of the type in which a description vector comprising at least one description coordinate defined using at least one description vector, is already known. a parameter calculated from a frequency spectrum of at least one sample segment of a signal representing the musical file. It should be noted that a conventional description vector is generally obtained after relatively long calculations, which can make the application of such a musical file description method long and laborious.

The object of the invention is in particular to remedy this drawback by providing a method for describing a musical file in which a description vector that can be obtained after simple and fast calculations is used, while maintaining a good quality of the description of the music file.

For this purpose, the object of the invention is a method for describing a musical file of the aforementioned type, characterized in that a first calculated parameter corresponds to a directing coefficient of a regression line of points of a space to two dimensions, the coordinates of these points being frequencies and amplitudes defining the frequency spectrum. This steering coefficient makes it possible to define a straight line forming an approximation of the shape of the frequency spectrum of the sample segment.

This steering coefficient makes it possible to obtain a coordinate of the description vector, called the oscillation of the frequency spectrum. It is obtained after simple calculations and satisfactorily describes the music file.

Preferably, a first descriptive vector description coordinate is an average of such master coefficients calculated over a plurality of samples. This first coordinate makes it possible to define an average approximate form of the spectrum of the signal. In a particular embodiment of the invention, the description vector comprises at least one description coordinate defined using a second calculated parameter corresponding to a speed of variation between two directional coefficients of two adjacent segments of the sample of predetermined identical durations. Indeed, knowing the directional coefficients associated with two identical adjacent segments, and knowing the duration of these segments, it is easy to calculate the speed at which the directional coefficient evolves between these two segments.

Preferably, a second descriptive vector description coordinate is an average of such master coefficient change rates calculated over a plurality of samples. This second coordinate, combined with the first coordinate, makes it possible to improve the description of the musical file.

In another particular embodiment of the invention, the description vector comprises at least one description coordinate defined by means of a third calculated parameter corresponding to a steering coefficient acceleration calculated using two transmission speeds. variation of the coefficients, each velocity being calculated between two adjacent segments of a triplet of adjacent segments of the sample.

Indeed, knowing the speeds associated with pairs of adjacent segments taken two by two in the triplet of adjacent segments, and knowing the duration of these segments, it is easy to calculate an acceleration of the steering coefficient between the three segments of the triplet.

Preferably, a third descriptive vector description coordinate is an average of such master coefficient variation accelerations calculated over a plurality of samples. This third coordinate, combined with the first and second coordinates, makes it possible to improve the description of the musical file. Note that it is still possible to improve the description of the music file. Indeed, optionally, at least one descriptive vector description coordinate is a standard deviation of parameters of the same nature calculated over several samples. Thus, the description vector is preferably defined in a six-dimensional space, its coordinates in this space being respectively the means and standard deviations of the driving coefficients, speeds and accelerations. This description vector with six description coordinates effectively describes the musical file.

In another embodiment, the description vector comprises at least one description coordinate defined using a calculated parameter corresponding to a derivative of order greater than two of the variation of the directing coefficient. In this case, the description vector may comprise any number of coordinates, each coordinate being calculated from a derivative of order greater than two of the variation of the steering coefficient. Generally, each sample is chosen according to a predefined distribution law in the musical file. Alternatively, each sample can be chosen randomly.

Preferably, the description vector of the musical file is incorporated into a description file associated with the musical file. Finally, in a particular embodiment, the description vector of the musical file is compared with a table of musical genres in order to determine a musical genre closest to the musical file.

The invention also relates to a method for comparing two music files with each other, characterized in that, each musical file being described according to a description method as defined above, the description vectors of the musical files are compared with each other in order to determine a similarity between these music files.

Preferably, a Euclidean distance is calculated between the respective description vectors of the two music files, these music files being considered to be similar if the calculated Euclidian distance is less than a predetermined threshold.

The invention also relates to a computer program for describing a musical file, characterized in that it comprises software instructions for the implementation of a description method as defined above. The invention also relates to a computer program for comparing two music files, characterized in that it comprises software instructions for the implementation of a comparison method as defined above.

The invention further relates to a server comprising a database in which music files are stored, characterized in that, each musical file being associated with a description vector obtained by a description method as defined above, the server comprises means for receiving a description vector, means for comparing this description vector with the description vectors of the musical files, and means for sending at least one file selected from a musical file whose vector of description is similar to the received description vector and a file with a link to such a music file.

The invention also relates to a terminal characterized in that it comprises means for storing a description vector obtained by means of a description method as defined above, means for sending the description vector to a server such that defined above, and means for receiving at least one music file sent by the server.

Finally, the invention relates to a device for describing a musical file, such as a terminal or a server, characterized in that it comprises means for implementing a description method as defined above. The invention will be better understood on reading the description which follows, given solely by way of example and with reference to the appended drawings in which:

FIG. 1 represents a signal representing a musical file;

FIG. 2 represents a frequency spectrum of a segment of the signal of FIG. 1;

FIG. 3 represents the steps of a description method according to the invention of the musical file represented by the signal of FIG. 1; and

FIG. 4 represents a computer system in which the description method of FIG. 3 is implemented. FIG. 1 shows a signal S, in the form of an amplitude / time curve. This signal S represents a musical file.

Also shown are samples E of the signal S. Each sample E has three sample segments si, s2, s3 adjacent of same predetermined durations. FIG. 2 shows points P of a two-dimensional space defined by a first axis ω of coordinates, representing frequencies, and a second axis A of coordinates, representing amplitudes. The points P define a frequency spectrum of a segment si, s2, s3 of a sample E of FIG.

Generally, these points P are obtained by means of a Fourier transform of the signal on the segment si, s2, s3 considered. Frequency spectra are defined in the same way for each segment si, s2, s3 of sample E.

These frequency spectra make it possible to calculate parameters making it possible to define description coordinates of a description vector of the musical file represented by the signal of FIG. 1. FIG. 2 also shows a regression line D of the points P obtained. using the Fourier transform of the signal on the segment si, s2, s3 considered.

FIG. 3 shows the steps of a description method of the musical file using this description vector. In a step 10, a sample E of the signal S is taken. Preferably, this sample is taken according to a predefined distribution law in the signal S, but it could alternatively be taken randomly.

During a step 20, for each segment si, s2, s3 of the sample E, a set of points P defining the frequency spectrum of the signal S on this segment si, s2, s3 is determined.

During a step 30, for each set of points P, a linear regression of these points P is carried out. This linear regression makes it possible to determine a regression line D, as represented in FIG. 2.

The directing coefficient of each line D thus determined forms a first calculated parameter. Each first calculated parameter corresponding to a segment si, s2, s3 of the sample E is then stored for later use.

In a step 40, for each pair of adjacent segments si, s2 and s2, s3 of the sample E, a second parameter calculated using the first calculated parameter of each segment of the pair of segments is determined. Each second calculated parameter corresponds to a speed of variation between two directing coefficients of two adjacent segments si, s2 (respectively s2, s3) of the sample E.

This speed is easily obtained by calculating a difference between the coefficients of the two adjacent segments, and dividing this difference by the duration of a segment. The sample E having three segments si, s2, s3 adjacent, thus determining two rates of variation of the director coefficient, each forming a second calculated parameter.

Each second calculated parameter is stored for later use.

In a step 50, a steering coefficient acceleration is determined, calculated using the two master coefficient variation rates calculated in step 40.

This acceleration is easily obtained by calculating a difference between these two calculated speeds, then dividing this difference by the duration of a segment.

The acceleration of director coefficient variations forms a third calculated parameter, which is stored for later use.

Thus, for the sampled sample E, six computed parameters have been stored, namely three guiding coefficients (first calculated parameters), two rates of variation of the guiding coefficient (second calculated parameters) and an acceleration of variation of the guiding coefficient ( third calculated parameter).

It is possible to repeat the steps 10 to 50 as many times as necessary, in order to determine parameters calculated for a sufficient number of samples to allow a good representation of the musical file. This sufficient number of samples depends for example on a size of the music file. In this case, therefore, probabilistic type sampling is performed, using a distribution law of the samples E making it possible to take only separate samples E two by two.

It should be noted that the number of sufficient samples is generally quite small. This rather small number reflects the fact that it is generally possible to recognize a musical genre by listening only a few seconds of a music file. This number of samples is preferably a parameter chosen during the implementation of the method.

In particular, the size of the set of samples taken is preferably smaller than the size of the signal representing the musical file, as can be seen in FIG.

It should be noted that a set of samples representing 10% of a musical file is generally sufficient to describe this musical file in a relevant way. Thus, it is not useful for all the samples to represent the entire music file. By reducing the number of samples taken, it reduces the time of implementation of the process. Thus, a set of samples representing between 10 and 50% of the musical file is preferably chosen, which constitutes an optimal compromise between a good quality of description of the musical file and a speed of implementation of the method.

When the number of samples taken is sufficient, we go to a step

60 during which an average is calculated, on all the samples E taken, of the first parameters calculated as well as an average, on all the samples taken, of the second calculated parameters, and an average, on all the samples taken, of the third calculated parameters.

Each of these averages forms a description coordinate of the description vector.

In this step 60, the standard deviations for each calculated parameter of the same nature are also calculated on all the samples taken. Each of these standard deviations forms a description coordinate of the description vector.

Thus, the description vector comprises six description coordinates.

During a step 70, the description vector is incorporated into a description file, called descriptor, associated with the musical file. Such a description vector can be stored very easily, its size being relatively small. Indeed, the size of such a description vector is for example of the order of 12 bytes.

In a step 80, the description vector of the musical file is compared with a table of musical genres for determining, according to the coordinates of the description vector, which is the musical genre closest to the musical file.

The music file has been described in such a way as to allow it to be classified in a musical genre.

The invention makes it possible to perform a method of comparing two music files with each other in the following manner. By comparing the description vectors associated respectively with the music files, determined according to the description method as described above, it is possible to determine whether the music files are similar or not, that is to say if they belong to the same genre musical.

For example, a Euclidean distance is calculated between the respective description vectors of the two music files. These music files are considered as being similar if the calculated Euclidean distance is less than a predetermined threshold.

In the case where the two compared music files contain the same musical piece, but of different qualities, their description vectors according to the invention are substantially identical. In this case, the Euclidean distance between these description vectors is substantially zero.

In the same way, in the case where the two compared music files contain the same musical piece, these musical files not having the same format, the Euclidean distance between the description vectors of these musical files is also substantially zero.

An exemplary implementation of the invention is described with reference to Figure 4, which shows a computer system designated by the general reference 100.

The computer system 100 includes a terminal 102, such as a server, a personal computer, a compact disc player, and so on.

The terminal 102 is capable of communicating with a server 104 and a device 105 for describing a musical file, for example a description server of a musical file, via a network 106, such as the Internet network. .

The description server 105 is provided with means 107 for implementing the previously described description method. For example, the implementation means 107 comprise means for storing a computer program for the description of a musical file, comprising software instructions for implementing the previously described description method and means for storage of a description vector obtained by this description method. Thus, the terminal 102 can obtain a description vector corresponding to a musical file by sending this musical file to the description server 105. This description server 105 then determines the description vector by means of the description method previously described and sends this vector of description. description at the terminal 102.

It will be noted that, alternatively, the terminal 102 could itself be a description device provided with means for implementing the previously described description method. In this case, the terminal 102 could itself determine a description vector corresponding to a musical file stored in this terminal 102.

The terminal 102 further comprises means 108 for sending to the server 104 a description vector obtained using the description method, and means 110 for receiving at least one musical file sent by this server 104. The server 104 includes a database 112 in which are stored music files and description vectors associated with these music files. These description vectors are obtained by means of the description method described above. The server 104 further comprises means 114 for receiving a description vector, means 116 for comparing this description vector with the description vectors of the music files, and means 118 for sending at least one file. , such as a musical file whose description vector is similar to the received description vector. The comparison means 116 are provided with means for storing a computer program for the comparison of two music files, comprising software instructions for implementing the comparison method previously described.

Thus, when a user of the terminal 102 wishes to obtain music files similar to a musical file stored in the terminal 102, he generates a description vector of this musical file using the description computer program, then sends the description vector to the server 104 using the sending means 108.

The server 104 receives this description vector using the reception means 114. The comparison means 116 then compare this description vector with the description vectors stored in the database 112 using the comparison method previously described. to determine at least one music file similar to the music file stored in the terminal 102.

When this similar musical file is determined, the file sending means 118 send this music file similar to the terminal 102. As a variant, the file sending means 118 could send a file containing a link to such a musical file. In this case, the user could use this link to download the music file.

It should be noted that the invention is not limited to the embodiment previously described.

Indeed, the description vector could include more or less description coordinates.

For example, additional description coordinates may be calculated from two-order derivatives of the variation of the steering coefficient. It should be noted that in order to calculate a derivative of order N of the variation of the directing coefficient, where N is any integer, it is necessary that each sample E taken has N + 1 segments. This derivative of order N is then calculated by iteration of calculations similar to the calculations of the aforementioned acceleration.

Claims

A method for describing a musical file, of the type in which a description vector comprising at least one description coordinate defined using at least one parameter calculated from a frequency spectrum of at least one sample segment (si, s2, s3) (E) of a signal (S) representing the musical file, characterized in that a first calculated parameter corresponds to a directing coefficient of a line (D) of regression of points (P) of a two-dimensional space, the coordinates of these points being frequencies (ω) and amplitudes (A) defining the frequency spectrum.

2. Method for describing a musical file according to claim 1, characterized in that the description vector comprises at least one description coordinate defined using a second calculated parameter corresponding to a speed of variation between two coefficients. director of two adjacent segments (si, s2, s2, s3) of the sample (E) of predetermined identical durations.

3. A method of describing a musical file according to claim 2, characterized in that the description vector comprises at least one description coordinate defined by means of a third calculated parameter corresponding to an acceleration of variation of the directing coefficient. calculated using two rates of variation of the coefficients, each velocity being calculated between two adjacent segments (si, s2, s2, s3) of a triplet of adjacent segments (si, s2, s3) of the sample ( E).

4. A method of describing a musical file according to any one of claims 1 to 3, characterized in that the description vector comprises at least one description coordinate defined using a calculated parameter corresponding to a derivative. of order greater than two of the variation of the guiding coefficient.

5. A method of describing a musical file according to any one of claims 1 to 4, characterized in that at least one descriptive vector description coordinate is an average of parameters of the same nature calculated on several samples (E ).

6. A method of describing a musical file according to any one of claims 1 to 5, characterized in that at least one descriptive vector description coordinate is a standard deviation of parameters of the same nature calculated over several samples ( E).

7. A method of describing a musical file according to any one of the preceding claims, characterized in that each sample (E) is chosen according to a predefined distribution law in the musical file.

8. A method of describing a musical file according to any one of the preceding claims, characterized in that the description vector of the musical file is incorporated into a description file associated with the music file.

9. A method of describing a musical file according to any one of the preceding claims, characterized in that the description vector of the musical file is compared with a table of musical genres in order to determine a musical genre closest to the musical file. .

10. A method of comparing two music files with each other, characterized in that, each musical file being described according to a description method according to any one of claims 1 to 9, vectors of description of the music files are compared with each other so to determine a similarity between these music files.

11. A method of comparing two music files according to claim 10, characterized in that a Euclidean distance is calculated between the respective description vectors of the two musical files, these music files being considered to be similar if the calculated Euclidean distance is less than a predetermined threshold.

12. Computer program for the description of a musical file, characterized in that it comprises software instructions for the implementation of a description method according to any one of claims 1 to 9.

13. Computer program for the comparison of two music files, characterized in that it comprises software instructions for the implementation of a comparison method according to claim 10 or 11.

14. Server (104) comprising a database (112) in which music files are stored, characterized in that, each musical file being associated with a description vector obtained by a description method according to any one of the claims. 1 to 9, the server comprises:

means (114) for receiving a description vector,

means (116) for comparing this description vector with the description vectors of the musical files, and

means (118) for sending at least one file chosen from a musical file whose description vector is similar to the received description vector and a file containing a link to such a musical file.

Terminal (102), characterized in that it comprises:

means for storing a description vector obtained by means of a description method according to any one of claims 1 to 9

means (108) for sending the vector to a server (104) according to claim 13, and

means (110) for receiving at least one musical file sent by the server (104).

16. Device for describing a musical file, such as a terminal (102) or a server (105), characterized in that it comprises means for implementing a description method according to any one of Claims 1 to 9.