JPH0990978A - Music constitution automatic extracting method of music information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To store all of the extracted music constitution information with a small amount of storage capacity by dividing music information into music constitutions based on the phrases detected from a detected voiced portion. SOLUTION: The music constitution automatic extraction is executed by a recording section 1 which records music information and stores the information in a music information file, an analysis file generating section 2 which processes the music information file and generates an analysis file and a music information analysis section 3 which analyzes the analysis file and extracts the music constitution. In other words, the section 1 generates the music information file. The music information file accumulated in the section 1 is processed in the section 2. Then, the analysis file is divided into phrases in the section 3. Thus, the voiced portion, which includes a voice or a melody, is detected and separated employing a voiceless portion from the music information that includes one or more than two acoustics. Then, the music information is divided into music constitutions based on the phrase which is one segment of a detected music.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music composition automatic extraction method for music information, and more particularly to a music information automatic composition extraction method for automatically extracting music composition from music information without score information.

[0002]

2. Description of the Related Art When extracting a music composition of music information, the music information is expressed in an electronic music description language using an electronic computer, and the beat, pitch, number of beats, and chord progression of the music information are obtained. It is possible to extract a phrase and music composition of music information based on a rule of information chord progression. The music composition consists of one or two or more phrases.
Generally, there are multiple song configurations.

Here, in order to cut out a characteristic portion called music digest information or chorus from the music information,
This music composition is used as a unit for cutting out. In this case, if the music information is expressed by a musical score or a music description language, the music composition can be easily extracted by dividing the music information for each rest.

[0004]

In the above music composition extraction method of music information, the music information must be re-expressed by an electronic music description language. It goes without saying that in order to do this, one must be familiar with electronic music description languages. Then, it takes a long time to re-express the original music information with the electronic music description language. In addition, this re-expression work also requires training to acquire the technique of auditory correspondence between the original music information for each timbre and the re-expression in the electronic music description language. However, there are various difficulties in automating this re-expression work.

Further, even if the music composition can be extracted from the music information re-expressed by the electronic music description language, the beat of the electronic computer is accurately carved in a fixed time. The beat produced by human performance is accompanied by a change in speed, and even in this respect, it is difficult to associate the original information with the extracted music composition. The present invention extracts music composition of music from original information of music information without requiring electronic re-description, and automatically extracts composition of music information to store all of extracted music composition information with a small storage capacity. It provides a method.

[0006]

A voiced part containing a voice or a melody in this music information is detected and separated from music information containing one or more sounds and voices by using an unvoiced part, and detected. An automatic music composition extraction method is constructed by detecting a phrase that is a segment of music from a voiced part and dividing the music information into music compositions based on the detected phrase.

[0007] The original song A and the original song A which are song songs
Both of the accompaniment A from which the singing voice is deleted are recorded as a music information file, and the recorded music information file is processed to create an analysis file of only the music parts of both the original song A and the accompaniment A. A music composition automatic extraction method of music information that analyzes only the analysis file is configured. Also, an automatic music composition extracting method for music information using music information recorded on a medium such as a compact disc or a record board is constructed.

Furthermore, an automatic music composition extracting method of music information using original karaoke as the original music A and the accompaniment music A is constructed.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention automatically extracts a music composition from music information recorded on a medium such as a commercially available compact disc or record board depending on the presence or absence of a voice portion composed of a singing voice, All the division point sequences of each music composition are accumulated. An embodiment of the present invention will be described with reference to FIG. FIG. 1 is a diagram for explaining the overall flow of a music composition automatic extraction method.

In FIG. 1, the music composition automatic extraction method is a recording unit 1 for recording music information and storing it in a music information file, and an analysis for processing the music information file to create an analysis file. File creation unit 2,
This is executed by the music information analysis unit 3 which analyzes the analysis file and extracts the music composition. Hereinafter, a case of extracting the music composition of the original music A will be described as an example. Original song A
Is a song song consisting of a song and other sounds. First, the recording unit 1 creates a music information file. The internal structure of the recording unit 1 is as shown in FIG. In the input section 11, the original song A which is a song entry, the original song A which is generally called original karaoke, the singing voice is deleted, or the voice or other sound recorded in the center phase of both left and right channels The accompaniment music A output from the audio device for erasing is recorded, and these are sampled and quantized by the file conversion unit 12, respectively, and accumulated as a music information file.

The music information file stored in the recording section 1 is processed in the analysis file creating section 2. The internal configuration of the analysis file creation unit 2 is as shown in FIG. The analysis file creation unit 2 processes the music information files of both the original song A and the accompaniment song A stored in the recording unit 1 into a form that can be analyzed. Since it is necessary to obtain the power difference between both songs in the analysis, both songs are synchronized. In order to synchronize both songs, first, in the silence detecting section 21, the music information files of both songs are sequentially arranged from the beginning to the rear.
A start point, which is a point having a value larger than a threshold value given as an argument for determining a silent portion, is detected. Similarly, the end point, which is a point having a value larger than the threshold value, is detected in order from the end to the front of the music information files of both songs. Next, in the music part cutout unit 22, the music information file is cut out from the start point to the end point detected by the silence detection unit 21 to analyze only the music parts of both the original song A and the accompaniment song A. Create a file.

Here, the music information analysis unit 3 divides the analysis file into phrases. The internal structure of the music information analysis unit 3 is as shown in FIG. The music information analysis unit 3 is an analysis unit that determines a start point and an end point that are break points at which the voice is separated, and a threshold value to which a ratio of unvoiced parts before the break points and voiced parts after the break points is given. The condition is that it is larger, and the threshold value is requested as an argument. This ratio is M
It will be described as.

First, in the starting point determining section 31, the correlation between the original music piece A and the accompaniment music piece A is calculated, and the point of both music pieces having the largest correlation value is regarded as the starting point to ensure the synchronization of both music pieces. Take Next, referring also to FIG. 5, the difference data calculation unit 3
2, the power of the beginning of the frame is calculated for each 200 msec frame length generally adopted in the voice recognition synthesis process for both songs, and the logarithm of the power of the beginning of the original song A and the power of the beginning of the accompaniment A are calculated. The logarithmic difference is calculated, and similarly, the difference between the head powers of the next frames is calculated, and the difference is calculated for each frame.

The unvoiced portion threshold value determination unit 33 is a determination unit that determines an unvoiced portion where only the accompaniment does not have a song, and the difference calculated by the difference data calculation unit 32 is larger than the threshold value given as the initial value. Are sequentially searched from the beginning difference, and the maximum difference within the range from the beginning to that point is determined as the unvoiced part threshold. The break point candidate detection unit 34 detects a break point at which a voiced part and an unvoiced part are switched. Search for a voiced part that is a point where the difference has a larger value than the threshold value determined by the unvoiced part threshold value determination part 33,
In the data range for the minimum length of 2 seconds that the interlude lasted before that point, which was examined from several sample songs, it was judged that it was a voiceless part, and conversely In the data range of 2 seconds in length, if the ratio of detection of a point judged as a voiced portion is larger than M,
That point becomes a voiced portion start point, that is, a breakpoint candidate.
Similarly, the ratio of detection of a point judged as a voiced part in the data range for 2 seconds before that point and a point judged as an unvoiced part in the data range for 2 seconds after that point is larger than M. A point that takes is also included in the breakpoint candidates. Separation point candidates are as indicated by vertical dotted lines in FIG.

The inappropriate candidate point detection unit 35 pays attention to the break point located at the start of the unvoiced part among the break point candidates,
The candidate for the break point of the sound-repellent stool or the consonant stool, which should not be detected as the break point, is excluded from the candidates. As the condition, the case where the difference before and after the point candidate changes as shown in FIGS. 12 and 13 is considered to be a correct point, and the point indicated by a solid arrow X in FIG. 7 is regarded as a point which does not satisfy this condition. exclude.

The cut point candidate detecting section 36 detects the points (hereinafter, referred to as cut points) S _{0 to} S _n at the break points of the phrase. In the inappropriate candidate point detection unit 35, by further doubling the threshold value of the maximum value of the slope, which is the reference of detection in the delimitation candidate detection unit 34, of the point sequences remaining in the array, the difference motion is further increased. I will list the big points as the cut points. Next, in the case of FIG. 12, since a long unvoiced part continues, the difference rises rapidly immediately before the rise, so that the rise of the difference is steep, but the four frames before that point are unvoiced. The portion that cannot be determined as a portion is excluded from the cut points, and the point indicated by the vertical dotted line in FIG. 8 is selected as the point that can be determined more accurately as the start of the voiced portion.

Referring to FIGS. 14 and 15, the singing point selecting section 37 is generally a singing point P ₀ -which is a candidate point for the starting point of each chorus, which is generally referred to as the first and second lyrics.
P _n is searched from the point sequence selected by the cut point candidate detection unit 36. First, it is considered that the chorus start point is a part that is judged to be unvoiced part for several frames from the singing start points P _{0 to} P _n , and the prelude part is continuously present in front. Therefore, one frame is set as the initial value of the length L of the prelude portion, and a point in which the unvoiced portion continues for L or more before the point is selected from the cut point sequence. Here, the chorus in one song is considered to be 8 or less based on the results of examination from several sample songs, and as a result of examining all points, when the starting point is 8 or more, L is short and the corresponding point is Since it is equal to a large number of selections, L is increased by 1 frame and the selection processing is performed in the same manner, and the selection processing is stopped when the start point becomes 8 or less. The selected singing point is as shown by the solid arrow in FIG.

The chorus start point determination unit 38 determines each chorus start point for one chorus and two choruses.
From the singing point sequence P _{0 to} P _n selected by the singing point selection unit 37, the starting points of 1 chorus and 2 choruses are selected. Usually the first of the singing point sequence
The singing point P ₀ is the start point of one chorus, and the second singing point P ₁ is the start point of two choruses. If there are several phrases at the beginning, this point is the start point of one chorus. Not 1 point. Therefore, by using the cut point sequences S _{0 to} S _n detected by the cut point candidate detection unit 36, the first point P ₀ and the second point P ₁ are the cut point sequence numbers as shown in FIG. If the first point S ₀ and the second point S ₁ match, it is determined that the prelude is composed of a few phrases and an interlude before the start of the chorus, and the second point P _{1 of} the start point sequence. Is the starting point of 1 chorus. The chorus start point is indicated by the solid arrow in FIG.

If the one chorus and the two choruses have the same structure, the music composition detecting section 39 has a condition that a cut point is detected at the same portion because a rest point is generated at the same portion on the score. The point that satisfies the condition is detected as the music composition start point. First, from the first point (X _i ) to the second point (X
_{i + 1)} to the data length of (L _1), 2 first point chorus (Y _j) from the second point (Y j _{+ 1)} to the data length of (L ₂₎ and compare, L ₁ = If L ₂ , the first and second points of each chorus
It is considered that the phrases sandwiched between the points are the same, and the first music composition is from the first point to the second point. Similarly, the second point is assigned to each of X _i and Y _j, and the third point is assigned to each of X _{i + 1} and Y _{j + 1} .
By substituting points, the data length L ₁ and the data length L ₂ are compared.

At this time, it is conceivable that there is a difference in singing style between the choruses, and the cut points are deviated. Therefore, when comparing the data lengths of both choruses, the deviation width W is set so that the data lengths are considered to be equal. When the initial value of the deviation width W is set to 0 and the absolute value | L ₁ −L ₂ | < W ... (a), it is considered that the same phrase exists. Here, if it is set that the deviation width W = 0, in this case, it is not considered that the same phrase exists unless the cut points appear at exactly the same place. Therefore, when the absolute value | L ₁ −L ₂ |> W (b) holds, the 1st chorus third point is substituted for X _{i + 1} . That is, L ₁ is set as the data length from the first point to the third point of one chorus and compared with L ₂ . In this way, when the expression (a) is satisfied, the same phrase exists from the first point to the third point of one chorus and the first point to the second point of the two chorus, which means that it is the first song composition. become. When the expression (a) is not established, X _{i + 1} is returned to the second point and Y _{j + 1} is incremented by 1. That is, the data length of the two choruses is changed and the comparison is similarly performed. Once the first song component is decided,
The end points of the first music composition of each chorus are set to X _i and Y _j , the next cut point candidates are set to X _{i + 1} and Y _{j + 1} , and the comparison and determination are performed in the same manner.

The division processing is advanced according to the shift width W, and when the division processing is completed, it is checked whether or not each chorus is divided into a number of divisions of 3 or more and 5 or less. The number of 3 divisions or more and 5 divisions or less is a statistical number of constituents obtained as a result of examining several sample pieces. If the number of corresponding divisions is within this range, the division processing ends. If the number of divisions does not fall within this range, the deviation width W is increased by one frame and division processing is performed again from the first point. Only the starting point of each music composition obtained by the above processing is stored as a composition point sequence. The result of the division into the musical composition is as shown by the solid double-headed arrow in FIG.

[0022]

As described above, according to the present invention, since the starting point for each music composition can be automatically obtained, by associating the music information used for music composition division with each point, By specifying the start point of the digest information, it is possible to automatically create the digest information according to the music composition without searching for the end point.

Even if it is necessary to sandwich the designated digest information and change the length of the digest information to be presented, since it is divided for each music composition,
Time is given priority, and musical composition is ignored, and there is no separation in the middle of a singing voice. Further, since only the start point of each music composition may be stored as a point sequence, the present invention can be applied even when it is desired to present music information with a small storage capacity.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the overall structure of a music composition division according to the present invention.

FIG. 2 is a diagram illustrating a flow of processing of a recording unit.

FIG. 3 is a diagram illustrating a processing flow of an analysis file creation unit.

FIG. 4 is a diagram illustrating a processing flow of a music information analysis unit.

FIG. 5 is a diagram showing changes in difference data.

FIG. 6 is a diagram showing detected demarcation candidate points.

FIG. 7 is a diagram showing inappropriate candidate points that cannot serve as a music composition break among the break candidate points.

FIG. 8 is a diagram showing cut point candidate points.

FIG. 9 is a diagram showing singing points.

FIG. 10 is a diagram showing a chorus start point.

FIG. 11 is a diagram showing a result of division into music compositions.

FIG. 12 is a diagram showing a change in difference data near a long rest.

FIG. 13 is a diagram showing changes in difference data in the vicinity of short rests.

FIG. 14 is a diagram showing a relationship between a singing start point and a chorus start point.

FIG. 15 is a diagram illustrating a processing flow of a singing point selection unit.

FIG. 16 is a diagram illustrating a processing flow of a music composition detecting unit.

[Explanation of symbols]

 1 Recording Section 11 Input Section 12 File Conversion Section 2 Analysis File Creation Section 21 Silence Detection Section 22 Music Part Extraction Section 3 Music Information Analysis Section 31 Starting Point Determining Section 32 Difference Data Calculating Section 33 Unvoiced Part Threshold Determining Section 34 Separation Candidate Detection Part 35 Inappropriate candidate point detection part 36 Cut point candidate detection part 37 Singing point selection part 38 Chorus start point determination part 39 Music composition detection part

Claims (4)

[Claims] 1. A voiced part containing a voice or a melody in this music information is detected and separated from music information containing one or more sounds and voices by using an unvoiced part, and music is detected from the detected voiced part. A method for automatically extracting music composition of music information, characterized by detecting a phrase that is a delimiter and dividing music information into music compositions based on the detected phrase. 2. The automatic music composition extraction method according to claim 1, wherein both the original song A which is a song entry and the accompaniment song A in which the singing voice is erased from the original song A are recorded as a music information file. Then, the recorded music information file is processed to create an analysis file of only the music part of both the original song A and the accompaniment song A, and the analysis file of only this music part is analyzed. Automatic configuration extraction method. 3. The automatic music composition extracting method for music information according to claim 1, wherein music information recorded on a medium such as a compact disc or a record board is used. A method for automatically extracting music composition of music information, which is characterized by the above. 4. The automatic music composition music extraction method according to claim 3, wherein original karaoke is used as the original music A and the accompaniment music A.