JP2002055695A - Music search system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a music search system which enables a user to efficiently search a music which the user desires to select. SOLUTION: A search key database 26 stores search keys for stored pieces of music, which simplified the change of musical interval in each phrase about a plurality of pieces of music each consisting of a plurality of phrases. When the sound collection of a melody, inputted by a singing voice or the like, is made by a microphone 20 and is converted into melodic data by a melodic recognition part 22, a search key preparation part 24 prepares a search key for an inputted piece of music which simplifies the change of musical interval about these melodic data. A search processing part 28 for the piece of music compares the search keys for the stored pieces of music, corresponding to each phrase of each piece of music with a search key for input piece of music, searches for the existence of a phrase, corresponding to the search keys for stored pieces of music having the same tendency as that of the search key for input piece of music about each of a plurality of pieces of music, and performs high-similarity determination to the piece of music, which has many phrases of the same tendency.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music retrieval system for retrieving a song desired by a user from a song database.

[0002]

2. Description of the Related Art Conventionally, in a karaoke apparatus, a unique identification number is added to each of a plurality of songs to be played, and a correspondence between the identification numbers and song names, artist names and the like is listed. A booklet is created, and the user can use this search booklet to check the identification number corresponding to the desired song, and input the identification number to the karaoke apparatus to play the desired song. It has become.

[0003] However, it has been difficult for a user to search for a desired song if the user does not know the song title or the like of the desired song, or if a search booklet is not at hand. Therefore,
A part of the melody included in the song is input by singing voice or instrumental sound, etc., and the input melody is converted into melody data (numerical data representing the temporal change of the pitch) by performing predetermined processing.
There has been proposed a music search system that searches for a desired song from a large number of song data recorded in a song database using the melody data.

For example, Japanese Unexamined Patent Publication No. 8-129393 discloses a technique for converting a melody input from a microphone into melody data and searching for a desired tune by comparing the melody data with melody data of a tune recorded in a tune database. Is described. Japanese Patent Application Laid-Open No. 9-293083 discloses that a melody input from a microphone is converted into predetermined pitch data (melody data) and rhythm data, and the melody is compared with the pitch data and rhythm data of the music recorded in the music database. A technique is described in which a desired music is searched for by calculating a similarity ratio between the two for each bar. In Japanese Patent Application Laid-Open No. 8-160975, a relative change in the pitch of the melody input from the microphone is obtained, and the desired change is searched for by comparing the relative change in the pitch of the song recorded in the song database. The method is described.

[0005]

By the way, the various retrieval methods disclosed in the above-mentioned publications extract music having a melody similar to the melody of the music to be searched from a music database. Are selected based on the degree of similarity. However, even if the degree of similarity is the same, there are cases where the melody of the phrase repeated in the song is similar and where the melody of a certain part happens to be similar by accident. Are determined to be the same. In practice, considering that phrases that are repeated many times are likely to remain in the listener's memory, evaluating these two songs in the same way would make it difficult for the user to efficiently search for the songs that the user wishes to select. Can not.

[0006] The present invention has been made in view of the above points, and an object of the present invention is to provide a music search system capable of efficiently searching for a song desired by a user to select. .

[0007]

In order to solve the above-mentioned problems, in the music search system of the present invention, for a plurality of tunes each composed of a plurality of phrases, a first variation in which a change in a pitch in each phrase is simplified. Characteristic data stored in the characteristic data storage means, and for the input voice data, second characteristic data in which a change in pitch is simplified is extracted by the characteristic data extraction means, and the extracted second characteristic data is extracted. And the first feature data corresponding to each phrase of each song stored in the feature data storage means are compared by the feature comparison means. Then, based on the comparison result by the feature comparison unit, the similarity determination unit checks the presence or absence of a phrase corresponding to the first feature data having the same tendency as the second feature data for each of the plurality of songs, and determines the same tendency. A high similarity determination is made for a song having a large number of phrases.

In general, the phrase (partial melody) that appears more frequently in a song tends to remain in the memory of the user, and the phrase that appears more frequently is often input by the user as a search phrase. . Therefore, by performing a higher similarity determination for a song having a larger number of phrases having the same tendency as the voice data corresponding to the phrase input by the user, the user is likely to remain in memory and listen to or sing. The tune to be considered is preferentially searched, and the tune desired by the user can be efficiently searched.

Further, the above-mentioned feature data extracting means extracts a vertex when observing a time change of the melody corresponding to the voice data, and represents another vertex whose pitch has changed from one vertex beyond a predetermined range. It is desirable to extract the data as a point, and to extract data representing a change in the pitch focused on these representative points as the above-mentioned second feature data. When the pitch does not change beyond the predetermined range, the pitch is not extracted as a representative point, so that a fine pitch change can be omitted. This effect can be eliminated.

The above-described feature data extracting means examines the amount of change in the pitch of the next representative point with reference to the pitch of one representative point, and performs an operation of classifying this change into one of a plurality of division ranges. It is desirable to repeat the process for each of the representative points and extract the classification result obtained for each of the representative points as the second feature data. Since the amount of change in pitch from one representative point to the next representative point is classified into any of a plurality of division ranges, the data amount of the second feature data can be reduced. The comparison process using the second feature data can be performed at high speed. Also,
Since the second feature data is extracted by a relative pitch change based on the pitch of one representative point, even when the key (tone) of the voice input by the user is shifted from the original music, Also, there is an advantage that similarity determination of music can be performed if the relative pitch changes match.

Preferably, the apparatus further comprises display means for displaying songs having a high degree of similarity in the order of the degree of similarity based on the result of the determination by the above-described similarity determination means. Thereby, the user can easily grasp the determination result.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A music search system according to an embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram illustrating a configuration of a music search system according to the present embodiment. The music search system shown in FIG. 1 includes a server device 1 and a terminal device 2 connected via a predetermined communication line.

The server device 1 is for transmitting music data based on a request from the terminal device 2 and includes a music database (DB) 10. Song database 1
No. 0 records song data corresponding to many songs in association with song names, artist names, and the like. In the present embodiment, the songs recorded in the song database 10 are basically vocal songs, that is, songs in which the main melody of the song (hereinafter, referred to as “main melody”) is expressed by a singing voice. Take into account. In addition, it is assumed that the song data in the present embodiment can be read at least separately from the main melody data corresponding to the main melody of the song and the accompaniment data corresponding to other accompaniments. . As such music data, for example, MIDI (Musical Instrument
ment Digital Interface) format.

The above-mentioned terminal device 2 is for inputting a melody by a predetermined method, searching for a desired tune based on the melody, and acquiring corresponding tune data. Unit 22, search key creation unit 2
4. It includes a search key database (DB) 26, a music search processing unit 28, and a display unit 30.

The microphone 20 converts a melody input by the user as a singing voice or a musical instrument sound into an analog electric signal. The melody recognition unit 22 converts the melody input by the user into digital melody data based on the electric signal output from the microphone 20. Specifically, the melody data described above is, for example, the microphone 20
After converting the electrical signal output from the digital signal into digital data, a well-known fast Fourier transform process is performed to extract a change in a frequency component corresponding to a lapse of time, that is, a pitch.

The search key creation unit 24 creates a predetermined search key that represents the melody input by the user in a simplified manner, based on the melody data output from the melody recognition unit 22. In the following description, a search key created based on the melody data output from the melody recognition unit 22 will be referred to as an “input music search key”. This input music search key corresponds to “second feature data”.

The search key creation unit 24 obtains music data recorded in the music database 10 in the server device 1, and simplifies the main melody included in the music based on the main melody data included in the music data. A predetermined search key is created. In the following description, a search key created based on the main melody data included in the music data recorded in the music database 10 will be referred to as a “recorded music search key”. The recorded song search key corresponds to “first feature data”. The details of the method for creating the above-described search keys (input music search key and recorded music search key) will be described later.

The search key database 26 stores the recorded music search keys created by the search key creation unit 24 in association with the music-specific information (for example, song names and artist names). The search key database 26 has
It is preferable to store recorded music search keys created corresponding to all music data recorded in the music database 10 in the server device 1.
Each time new song data is called from the server device 2 to the terminal device 1, a recorded song search key may be created corresponding to the song data. Alternatively, a recorded music search key may be created in advance corresponding to a large number of music data and stored in the search key database 26.

The music search processing section 28 includes a search key creation section 24
Of the input music search key created by the melody recognition unit 22, and compares the obtained input music search key with the recorded music search keys stored in the search key database 26 to determine the similarity. Search for a tune corresponding to the melody input by the user. The display unit 30 displays a result of the search performed by the music search processing unit 28 and the like.

The above-mentioned search key database 26 is used as the feature data storage means for the melody recognition unit 22 and the search key creation unit 2.
Reference numeral 4 corresponds to a feature data extraction unit, the song search processing unit 28 corresponds to a feature comparison unit and a similarity determination unit, and the display unit 30 corresponds to a display unit. The music search system according to the present embodiment has such a configuration, and its operation will be described in detail below.

(A) Procedure for Creating Recorded Song Search Key FIG. 2 is a flowchart showing an operation procedure of the music search system when creating a recorded song search key based on song data recorded in the song database 10. Mainly search key creation unit 2
4 describes the operation performed.

First, the search key creation unit 24 acquires music data from the music database 10 in the server device 1 and extracts main melody data (step 100). FIG. 3 is a diagram showing an example of the main melody data included in the music data acquired from the music database 10, in which the horizontal axis corresponds to the elapsed time and the vertical axis corresponds to the pitch.

In this embodiment, one octave is divided into 12 sections for each semitone, and the semitone is the minimum unit of the pitch difference. In the following, for the sake of simplicity, the pitch difference between semitones is expressed as "the pitch difference is 1". That is, the pitch difference of one octave is “the pitch difference is 1
2 ".

Next, the search key creation unit 24 divides the main melody into a plurality of phrases (step 101). In the present embodiment, a partial melody included in each section when the main melody is divided into several sections is called a “phrase”. FIG. 4 is a diagram illustrating a phrase. In the example shown in the figure, a state where the main melody shown in FIG. 3 described above is divided into four phrases is shown. As for the method of dividing a phrase, for example, a person who produces song data including main melody data includes predetermined segment information indicating a segment of a phrase in song data in advance, and divides based on this segment information. Various methods are conceivable, such as a method, a method of automatically dividing every predetermined number of measures, and a method of dividing by detecting a rest of a predetermined length (a part where the melody is interrupted).

When the main melody is divided into a plurality of phrases,
Next, the search key creation unit 24 extracts the vertices of the main melody data for each phrase, and stores the position (time and pitch) of each vertex (step 102). FIG. 5 is a diagram specifically explaining the vertices of the main melody data. As an example, FIG.
A state in which vertices are extracted based on the main melody data shown in FIG. 4 and the like described above is shown. In FIG.
Each position indicated by (21) indicates a vertex of the main melody data. Specifically, in the present embodiment, a position where the main melody data is convex upward or a position where the main melody data is convex downward is called a vertex. For example, in FIG. 5, when moving from the position (1) to the position (2), the pitch is lowered,
When the pitch changes from the position (2) to the position (3), the pitch is raised. In this case, the position (2) is a “position that is convex downward”. Extracted as vertices. Similarly, when the position changes from the position (2) to the position (3), the pitch increases, and when the position changes from the position (3) to the position (4), the pitch continues to decrease. In this case, the position (3) is “a position where the melody data is convex upward”, and this position is extracted as a vertex. The other vertices are similarly extracted.

When the vertices are extracted, the search key creation unit 24 sets an error range E relating to the pitch difference, examines the vertices extracted in the above-described step 102 in order from the top, and determines the pitch difference between the vertices. Is extracted as a representative point, and the vertices that change beyond the range of E / 2 are extracted (step 103).

FIG. 6 is a diagram specifically explaining the representative points. In FIG. 6, for example, if the error range is set to E = 2, vertices whose pitch difference exceeds 1 with respect to the vertex closer to the beginning (vertex of interest) are extracted. Therefore, in the first phrase, when the vertex of interest is the vertex (1), the next vertex (2)
Since the pitch difference between and the vertex (1) is 3, the vertex (2) is extracted as a representative point.

If the target vertex is vertex (2), the pitch difference between the next vertex (3) and vertex (2) is 1, and vertex (3) is not a representative point. If the target vertex is vertex (3), the pitch difference between the next vertex (4) and vertex (3) is 5, so vertex (4) is extracted as a representative point. Similarly, assuming that the target vertex is vertex (4), since the pitch difference between the next vertex (5) and vertex (4) is 4, vertex (5) is extracted as a representative point. To summarize the above results, in the first phrase, the vertex (1),
Four (2), (4), and (5) are extracted as representative points.

In the same manner as described above, the vertices (6), (7), (9), (10),
Five of (11) are extracted as representative points, three vertices (12), (13), and (15) are extracted as representative points in the third phrase, and vertices (16),
Five points (17), (19), (20), and (21) are extracted as representative points.

When a representative point is extracted from each phrase, the search key creating unit 24 creates predetermined branch data based on the degree of the pitch difference between the extracted representative points (step 104). Here, first, the threshold value W relating to the pitch difference is set to a predetermined value, and based on the set threshold value W, a certain representative point and a next representative point (forward on the time axis) for each phrase. Are classified.

Specifically, in the present embodiment, the relation between the representative points is as follows: (1) When the pitch difference changes beyond + W,
(2) When it is within the range of 0 to + W, (3) When it is within the range of 0 to -W, (4) When it is changed beyond -W,
It is classified into four. In this case, “+” added before the above-mentioned threshold value W indicates that the pitch of the next representative point is raised with respect to a certain representative point, and “−”
Indicates that the pitch of the next representative point is lowered with respect to a certain representative point.

FIG. 7 is a diagram for explaining the tree data. FIG. 7A shows the first case where threshold value W = 5.
Shows the contents of the binary tree data obtained in correspondence with the phrase of As shown in the figure, four branches are prepared in the branch tree data of the present embodiment, and correspond to the above-mentioned four types of relations between the representative points classified based on the value of the threshold value W. Any one branch is selected. For example, in the first phrase, the pitch difference between the representative point (1) and the representative point (2) is -3, which corresponds to the above (3) "when the value is within the range of 0 to -W". Therefore, the third branch from the top is selected. Similarly, since the pitch difference between the representative point (2) and the representative point (4) is -4, the third branch from the top is selected, and the difference between the representative point (4) and the representative point (5) is obtained. Is +4, the second branch from the top is selected. When these selected branches are connected and drawn, as shown by the solid line in FIG. , The first phrase is represented in a simplified manner by the binary tree data having a pattern of “the third branch from the top, the third branch from the top, and the second branch from the top”.

Also, for the second to fourth phrases,
Branch tree data is obtained in the same manner as in the case of the first phrase described above. FIG. 7 (B) is the tree data obtained corresponding to the second phrase, FIG. 7 (C) is the tree data obtained corresponding to the third phrase, and FIG. 7 (D) is the fourth tree data. , Respectively, respectively.

When the branch tree data corresponding to each phrase is obtained, the search key creation unit 24 checks the patterns of these branch tree data and checks the number of appearances of the same pattern (step 105). Here, in the branch tree data, the first branch from the top is “branch 1”, the second branch from the top is “branch 2”, the third branch from the top is “branch 3”, and the fourth branch from the top is “branch 3”. "Branch 4"
In this case, the patterns of the branch data corresponding to the first to fourth phrases shown in FIG. 7 are expressed as follows.

First phrase: (branch 3-branch 3-branch 2) Second phrase: (branch 3-branch 3-branch 1-branch 3) Third phrase: (branch 2-branch 2) 4th Phrase: (branch 3-branch 3-branch 1-branch 3) When examining the patterns of the first to fourth phrases, the pattern of the branch tree data corresponding to the first phrase and the pattern of the third phrase are determined. It can be seen that each pattern of the branch tree data appears only once in the music. Further, since the content of the pattern of the branch tree data corresponding to the second phrase and the pattern of the branch tree data corresponding to the fourth phrase match, the number of appearances of the pattern of the branch tree data is twice. It turns out that.

Next, the search key creation unit 24 classifies the branch tree data based on the number of appearances of each pattern, and creates a recorded music search key using the classified branch tree data (step 106). FIG. 8 is a diagram for explaining a recorded music search key. FIG. 8A shows a recorded music search key created based on the binary tree data corresponding to the first and third phrases in which the number of appearances of the pattern is one. FIG.
As shown in (A), by combining the binary tree data corresponding to the first and third phrases, a pattern (branch 3-branch 3-branch 2) corresponding to the first phrase and a third Tree data including a pattern of (branch 2-branch 2) corresponding to the phrase is obtained, and this is one of the recorded music search keys.

FIG. 8B shows a recorded music search key created based on the binary tree data corresponding to the second and fourth phrases in which the pattern appears twice. As shown in FIG. 8B, based on the tree data corresponding to the second and fourth phrases, tree data having a pattern of (branch 3-branch 3-branch 1-branch 3) is obtained. This is also one of the recorded song search keys.

When the recorded song search keys are obtained, the search key creating unit 24 determines, for each of the obtained recorded song search keys, according to the number of appearances of the pattern of the branch tree data from which these were created. Then, a predetermined weighting coefficient is added (step 107). In the recorded music search key shown in FIG. 8 described above, for example, “1” is added as a weighting factor to the recorded music search key shown in FIG.
"2" is added as a weighting coefficient to the recorded music search key shown in FIG.

Even when three or more recorded song search keys are obtained, weighting factors are set appropriately according to the number of appearances of the pattern of the branch tree data from which each recorded song search key was created. And be added. next,
The search key creating unit 24 stores the search key in the search key database 26 in association with the information such as the title of the recorded music after the weighting coefficient is added (step 108).

Through the above-described series of procedures, the creation of the recorded music search key is completed. Also, by changing the error range E and the threshold value W to various values and repeating the procedure of steps 103 to 108 shown in FIG. A key is created. For example, in the example described above, the error range is E =
2. Although the threshold value is set to W = 5, under the condition that one or both of the error range and the threshold value are changed,
About ten types of recorded music search keys are created for one music.

(B) Procedure for Retrieving a Desired Song Next, a procedure for retrieving a song desired by the user based on the melody input by the user will be described.
FIG. 9 is a flowchart showing an operation procedure of the music search system when searching for a desired song.

First, the melody recognition unit 22 determines whether or not a melody is input by a singing voice, a musical instrument sound, or the like (step 2).
00), when a melody is input, an affirmative judgment is made, and the input melody is converted into melody data (step 20).
1). When the input melody is converted into melody data, the search key creation unit 24 extracts the vertices of the melody data,
The position of each vertex is stored (step 202). Next, the search key creator 24 sets an error range E relating to the pitch difference, examines the vertices extracted in step 202 described above in order from the top, and determines that the pitch difference between the vertices exceeds the range of E / 2. A changing vertex is extracted and is extracted as a representative point (step 203).

When the representative points are extracted, the search key creation unit 2
4 obtains branch tree data based on the degree of the pitch difference between the extracted representative points, and creates an input music search key based on this (step 204). Specifically, the melody input by the user in step 200 described above is
Generally, it is considered that the melody is a part of the main melody included in the song, that is, a phrase. Therefore, in step 204, the whole of the input melody is regarded as one phrase. Therefore, in step 204, the branch tree data obtained by performing the predetermined processing directly serves as an input music search key.

The procedure for extracting the vertices and the representative points of the melody data and the procedure for creating the branch tree data based on the representative points are the same as the procedures for creating the recorded music search key described above. Here, detailed description is omitted. As described above, in the present embodiment, the recorded music search key and the input music search key are basically created by the same algorithm, so that the target music can be more reliably searched.

FIG. 10 is a diagram showing an example of the input music search key created in step 204. In this figure, as an example, an input music search key when the user inputs the second phrase or the melody corresponding to the fourth phrase in the main melody shown in FIG. 4 and the like described above is shown. If the user inputs the melody corresponding to the second phrase or the fourth phrase with some accuracy, as shown in FIG. 10, (branch 3-branch 3-branch 1-branch 3)
Is obtained, and this is used as an input music search key.

Next, the song search processing unit 28 receives the created input song search key from the search key creation unit 24 via the melody recognition unit 22, and stores the input song search key and the search key database 26. A search is made for the songs that include the recorded song search key that matches or is similar to the input song search key (step 205), and the search keys (input song search key and recorded song search) are compared. The matching rate is calculated in consideration of the degree of similarity of the key and the weighting coefficient added to the recorded music search key (step 206).
Specifically, in the present embodiment, the pattern of the branch tree data corresponding to the input music search key and the pattern of the branch tree data corresponding to the recorded music search key are compared, and only one branch point of the branch tree data is found. If they are different, it is determined that each search key is “similar”.

FIG. 11 is a diagram showing an example of the calculation result of the precision. The songs A to D shown in the figure are examples of songs that include a recorded song search key that is similar or coincident with the input song search key. It is assumed that an input music search key (branch 3-branch 3-branch 1-branch 3) shown in FIG. 10 is considered as an input music search key.

The song A shown in FIG. 11 is the song described with reference to FIG. 8 and the like described above, and the included song search key (weight 3 = branch 3—branch 1—branch 3) with the weighting coefficient = 2 is the input music search key. Matches. Also, for the music B, the recorded music search keys (the branch 3—the branch 3—the branch 1—the branch 3) with the weighting coefficient = 1 coincide with the input music search key. Similarly, the song C has a recorded song search key (weight 3-branch 3-branch 1-branch 2) with a weighting coefficient = 2 similar to the input song search key. The song D has a recorded song search key with a weighting factor of 1 (branch 3-branch 3-branch 2-branch 3) similar to the input music search key.

In this embodiment, the matching rate is expressed as a percentage. First, based on the similarity between the input music search key and the recorded music search key, (a) no deduction for perfect match, ( b) In the case of similarity, a penalty point of 10% is deducted from the precision rate. Further, based on the weighting coefficient, the point with the largest weighting coefficient is regarded as no deduction, and a deduction of 5% is performed every time the weighting coefficient decreases by one.

More specifically, when comparing the input music search key and the recorded music search key with the weighting coefficient = 2, there is no penalty related to the degree of similarity for the music A because it is a perfect match (that is, the penalty is 0%). ), The weighting coefficient = 2 (the largest weighting coefficient in the music piece A), so that there is no penalty for the weighting coefficient, and thus the precision is calculated as 100%. Also, when comparing the input music search key and the recorded music search key with the weighting coefficient = 1 for the music B, there is no penalty related to the degree of similarity because it is a perfect match, but since the weighting coefficient = 1, the weighting coefficient is 5%. Is performed, and the precision is therefore calculated to be 95%.

Similarly, when the input music search key and the included music search key with the weighting coefficient = 2 are compared for the music C, they are similar, so that the degree of similarity is reduced by 10%, and the weighting coefficient = 2. Therefore, there is no penalty for the weighting factor, and thus the precision is calculated as 90%.
Similarly, for song D, when the input song search key is compared with the included song search key with the weighting coefficient = 1, the similarity is deducted by 10% with respect to the degree of similarity. Is deducted by 5%, and thus the precision is calculated to be 85%.

Here, looking at the calculation result of the matching rate shown in FIG. 11, both the music A and the music B include the recorded music search key that completely matches the input music search key.
It can be seen that the calculation result of the relevance ratio is higher in the case of, because the weighting coefficient added to the recorded song search key is larger. Further, both the songs C and D include a recorded song search key similar to the input song search key, but the song C has a larger weighting coefficient added to the recorded song search key because the song C has a larger weighting coefficient. It can be seen that the calculation result of the precision is higher. A song search key with a large weighting factor corresponds to a phrase that appears more frequently in the song, and since such a phrase that appears more often is heard more than once, it is remembered and input by the user. Since it is highly likely to be given as a song search key, the relevance is calculated in consideration of the weighting coefficient, so that the relevance of a song that the user is likely to want to select is calculated to a higher value. be able to.

Further, as described above, in this embodiment, 1
Since several (for example, 10) recorded music search keys are created for the music by changing the values of the error range E and the threshold value W, the music search processing unit 28 A final matching rate for one song is calculated by integrating the calculation results of the matching rates using all the search keys (step 207). Specifically, for example, if ten types of recorded song search keys are prepared for the above-described song A, the matching rate based on a comparison between each of the recorded song search keys and the input song search key is determined. Since ten calculation results are obtained, these calculation results are added, and the addition result is divided by 10, thereby obtaining the final calculation result of the precision.

After that, the song search processing section 28 extracts the songs including the included song search keys whose final relevance ratio calculation result is equal to or more than a predetermined value (for example, 70% or more) and extracts them in a predetermined order (for example, The song name, the artist name, and the like and the calculation result of the matching rate are displayed on the display unit 30 in the order of the highest matching rate (step 208). Then, when the user specifies a desired song from the one or more candidate song lists displayed on the display unit 30 using an operation unit or the like (not shown), the song search processing unit 28 sends the song to the server device 1. In response to this request, a song data transmission request is output, and the song data is sent from server device 1 in response to the request.

As described above, in the music search system according to the present embodiment, for a plurality of songs each composed of a plurality of phrases, when creating a recorded song search key in which the change in the pitch in each phrase is simplified, , A larger weighting coefficient is added to a recorded music search key created corresponding to a phrase that appears more frequently. Then, when the user inputs a melody to search for a song, an input song search key that simplifies the change in pitch is created correspondingly, and the input song search key and each of the plurality of songs are created. The similarity is determined by comparing a correspondingly prepared song search key with a weighting factor added to the song search key. Therefore, a song containing many phrases having the same tendency as the melody input by the user, that is,
A song having a higher possibility of being desired by the user has a higher determination result of the similarity determination, and the user can efficiently search for a song desired to be selected.

The present invention is not limited to the above-described embodiment, and various modifications can be made within the scope of the present invention. For example, in the embodiment described above,
Each search key is created by using the branch data in which four branches are set. However, the number of branches to be set in this branch data may be at least two, and more branches may be set in reverse. . The smaller the number of branches to be set, the smaller the data amount of each search key, and the more branches, the more detailed the characteristics of the phrase can be expressed.

In the above-described embodiment, as a criterion for judging the degree of similarity between the input music search key and the recorded music search key, a case where the pattern of the branch tree data differs by one place is “similar. However, the criterion for determining the degree of similarity is not limited to this, and may be determined based on another criterion. Specifically, for example, when one part of the pattern of the branch tree data is different, a deduction of 10%, 2
If the points are different, the deduction may be set according to the increase or decrease of the points with different patterns, such as 20%,....

In the above-described embodiment, the minimum unit of the pitch change in the main melody data and the like is a semitone.
The present invention is not limited to this, and the minimum unit of the pitch change may be set to a value other than a semitone. Further, in the above-described embodiment, several types of recorded music search keys are prepared for each music, but it is not always necessary to prepare several types, and at least
It suffices if one kind of recorded music search key is prepared for one music.

In the above-described embodiment, the description has been made that the songs recorded in the song database 10 are basically vocal songs, but the songs recorded are not limited to this. Instead, a song in which the main melody is represented by an instrumental sound other than the singing voice may be recorded. Specifically, for example, a song belonging to a genre such as classical music or jazz can be considered. In this case, a melody that most impresses the song is selected as a main melody, and corresponding main melody data is prepared. It is good.

[0060]

As described above, according to the present invention, the higher the number of phrases having the same tendency as the phrase corresponding to the input voice data, the higher the similarity determination is made, so that the similarity determination remains in the memory. As a result, songs that the user himself wants to listen to or want to sing are searched with priority, and songs that the user wants to select can be efficiently searched.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a configuration of a music search system according to an embodiment.

FIG. 2 is a flowchart showing an operation procedure of the music search system when creating a recorded music search key based on music data recorded in a music database.

FIG. 3 is a diagram showing an example of main melody data included in song data acquired from a song database.

FIG. 4 is a diagram illustrating a phrase.

FIG. 5 is a diagram specifically illustrating a vertex of main melody data.

FIG. 6 is a diagram specifically illustrating a representative point.

FIG. 7 is a diagram illustrating branch tree data.

FIG. 8 is a diagram illustrating a recorded music search key.

FIG. 9 is a flowchart showing an operation procedure of the music search system when searching for a desired song.

FIG. 10 is a diagram showing an example of an input music search key.

FIG. 11 is a diagram illustrating an example of a calculation result of a precision.

[Explanation of symbols]

 Reference Signs List 1 server device 2 terminal device 10 music database (DB) 20 microphone 22 melody recognition unit 24 search key creation unit 26 search key database (DB) 28 music search processing unit 30 display unit

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI theme coat ゛ (reference) G10L 15/00 G10L 3/00 551G F-term (reference) 5B075 ND14 PQ04 PR06 QM08 5D015 AA06 HH04 5D108 BC01 BC17 BG06 5D378 KK01 MM97 QQ01

Claims (4)

[Claims] 1. A feature data storage means for storing first feature data in which a change in pitch in each phrase is simplified for a plurality of songs each including a plurality of phrases; Feature data extraction means for extracting second feature data whose change has been simplified; the second feature data extracted by the feature data extraction means; and each phrase of each song stored in the feature data storage means A feature comparison unit that compares the first feature data with the first feature data corresponding to the second feature data;
The presence / absence of the phrase corresponding to the first feature data having the same tendency as that of the first feature data is checked for each of the plurality of songs, and a high similarity determination is performed for a song having the same tendency and a large number of phrases. A music search system comprising: a similarity determination unit. 2. The method according to claim 1, wherein the characteristic data extracting unit extracts a vertex when observing a time change of a melody corresponding to the voice data, and a pitch changes from one vertex beyond a predetermined range. A music search system, wherein other vertices are extracted as representative points, and data representing a change in pitch focused on these representative points is extracted as the second feature data. 3. The characteristic data extracting unit according to claim 2, wherein the characteristic data extracting means checks a change amount of a pitch of the next representative point with reference to a pitch of one representative point, and determines the change amount in any of a plurality of division ranges. A music search system, wherein an operation of classifying a crab is repeated for each of the representative points, and a classification result obtained for each of the representative points is extracted as the second feature data. 4. The music according to claim 1, further comprising: display means for displaying songs having a high degree of similarity in order of the degree of similarity based on a result of the determination by the similarity determination means. Search system.