JPH07117833B2 - Musical note similarity calculator - Google Patents

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 using note sequence cues containing an error, and more particularly, a note sequence similarity calculation for calculating a similarity between a clue note sequence and a music note sequence. Regarding the device.

[0002]

2. Description of the Related Art In a music search using note string information as a clue, it is necessary to check which music contains a note string which is a clue to the search. For this purpose, it is necessary to compare and collate the musical note sequence, which is a clue of the search, and the musical note sequence of the musical piece to be searched. In particular, when the clue of the search is inaccurate, the clue and the musical note sequence of the musical piece are not always exactly the same. In such a case, not only is it determined whether or not there is a match between the cue note sequence and the musical note sequence, but also the degree of similarity (similarity) between the cue note sequence and the musical note sequence is calculated. However, it is necessary to perform processing such that the music having the maximum similarity is used as the search result.

Conventionally, in order to obtain the similarity between note strings,
There is known a device that pays attention to the notes included in the note string and obtains the similarity between the note strings from the comparison result between the notes. One example is "melody search-searching music DB from humming-", Proceedings of the 43rd IPSJ National Convention, IPSJ, 3M-5, October 1991 (hereinafter,
Reference 1) ”.

Hereinafter, the inter-note-note sequence similarity calculation device described in Document 1 will be described with reference to the drawings. Figure 8
[Fig. 6] is a block diagram showing an operation of the inter-note-note sequence similarity calculation device described in Document 1. With reference to FIGS. 8 and 9A, the operation of the inter-note sequence similarity calculation device described in Document 1 will be described using a specific example.

When calculating the similarity between note strings of note string 1 and note string 2 of FIG. 9A, the note string 1 is input by the input means 1.
And note sequence 2 is given. Next, note correspondence calculation means 2
Determines the correspondence of notes to be compared between note strings. For example,
When 2 <i <5, the correspondence between the i-th note of note sequence 1 and the i-th note of note sequence 2 is obtained. The note coincidence calculation means 3 obtains a comparison result between notes based on the correspondence. For example, by matching the note coincidence based on the correspondence
The number of notes whose pitch difference between the current note and the immediately preceding note matches
It is defined as At this time, with the note immediately before each note
Compare the pitch difference between note sequence 1 and note sequence 2
I get the value 3 which is the number of notes I have
Become. The note string deformation point deduction calculation means 11 finds a score deduction of the similarity due to the deformation when the note string is deformed, but since the note string to be compared is not deformed in the comparison between the note string 1 and the note string 2, The score deduction is 0. The similarity calculation means 5 outputs the value 3 which is the sum of the value 3 of the note coincidence and the value of the note string deformation deduction as the similarity between the note strings to the output means 6.

Further, in FIG. 9B, when the note sequence similarity between the note sequence 1 and the note sequence 3 is obtained, the note coincidence is 3 and the note sequence similarity is calculated by the same calculation as above. Is 3.

[0007]

When retrieving a musical piece from an inaccurate note sequence clue, the correct portion of the musical note sequence as a clue becomes a clue for selecting a musical piece.
The inputted note sequence clues are part of the music that the user who is trying to search remembers, but the part of the clue note sequence where the memory is ambiguous tends to be inaccurate. For example, when the memory of the beginning or end of a clue is ambiguous, there are many cases where this part is an incorrect clue. In this way, the wrong part of the note sequence cues is often unevenly distributed in a specific place,
It is rarely dispersed throughout. On the contrary, the longer the non-erroneous portion of the clue note sequence is, the more effective the portion of the musical piece selection becomes. From the above, when the clue and the music are similar for a long time, it is necessary to calculate the similarity so that the similarity between the clue and the music is high.

However, in the conventional sequence of notes between the similarity calculation device described in Document 1, a similarity between the sequence of notes, sound
It is calculated only by deductions due to the number of modification times of the musical note sequence for comparison with the note coincidence . Further, the note coincidence is evaluated only by the pitch difference from the immediately preceding note, and therefore does not have information about the similarity of the preceding and following notes. That is, in the inter-note-note-similarity calculating device described in Document 1, there is a problem in that the degree of continuity of similar portions is not reflected in the similarity, because the degree of similarity is calculated according to the ratio of errors in the entire note string. is there.

This problem will be specifically described with reference to FIG. The note sequence 2 is significantly different from the note sequence 1 in the middle part. On the other hand, note sequence 3 is different from note sequence 1 only in the last note. However, in the note string similarity calculation device described in Document 1, the note string 2 and the note string 3 both have a note string similarity of 3 with the note string 1, and which note string is the note string 1 Could not be determined to be similar to.

The present invention solves the above-mentioned problems, and an object of the present invention is to make note sequence similarity between note sequences in which continuous similar portions included in an inaccurate note sequence input as a clue can be utilized as effective clues. To provide a degree calculator.

[0011]

SUMMARY OF THE INVENTION A first aspect of the present invention is to arrange musical notes.
To compare and collate the musical note sequences represented as
, Using the comparison result of each note included in the note string,
In the similarity calculation device between note sequences for calculating the similarity between note sequences , the correspondence of the notes to be compared and collated between the note sequences is determined by the entire note sequence.
Based on the note correspondence calculation means found by the body and all of the correspondence
From the result of comparison and matching between notes,
Note matching degree calculation means for determining the degree of note matching based on the response , and the note sequence between the note strings based on the correspondence.
If the marks match in succession, the degree of similarity in
It is characterized by further comprising a similarity correction amount calculation means for obtaining the same similarity correction amount, and a similarity calculation means for calculating a similarity between note strings from the note coincidence and the similarity correction amount.

A second aspect of the present invention is the inter-note string similarity calculating device according to the first aspect of the present invention, wherein the similarity correction amount calculating means includes the pitch value of the notes included in the note string based on the correspondence and the notes. Plural-preceding pitch difference calculating means for obtaining a plural-preceding pitch difference value that is a difference from the pitch value of the plural-preceding preceding note, A high-difference coincidence calculation means is provided.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the first invention. The present embodiment is based on the input means 1 to which two note strings for calculating the degree of similarity are input, the note correspondence calculating means 2 for obtaining the correspondence between the notes to be compared between the two note strings, and the note correspondence. Note matching degree calculation means 3 for obtaining the degree of matching between notes, similarity degree correction amount calculating means 4 for obtaining the degree of similarity correction according to the degree of similarity of consecutive notes based on the correspondence of the notes, and note agreement The similarity calculation means 5 calculates the similarity from the degree and the similarity correction amount, and the output means 6 outputs the similarity.

The input means 1 is supplied with note strings A and B for which the degree of similarity is calculated. As input means, keyboard input of keyboard instruments, note input using a character keyboard,
MIDI input, which is a music data transfer standard, input using a mouse / touch panel / trackball / pen input device, input from a music storage device, and a note string is generated from an acoustic signal obtained by a microphone or the like (input ( For example,
"Personal computer music system-Automatic transcription of singing voice-, Proceedings of the 35th Annual Conference of IPSJ,
Information Processing Society of Japan, 5Ff-5, 1987 "). Of course, the two note strings may be input by different input devices. Each of the two note strings input by the input means is a string in which a plurality of notes are arranged.
The note sequence may include rests at the beginning, between notes, between notes, and the like. In addition, each note has a pitch value representing the pitch of each pitch as an attribute value, or a pitch value indicating the length of the pitch, and may have both the pitch value and the pitch value as necessary. . Each rest may have a rest length attribute value indicating the length of the rest.

The note correspondence calculation means 2 includes input means 1 to 2
It receives two note sequences and finds the correspondence of the notes to be compared between the received note sequences. As an example of how to find the correspondence between notes, when i is a natural number, the correspondence between the i-th note from the beginning of the note string, the correspondence between the i-th note from the end of the note string, and the note string I from the beginning
There are ones that correspond to the th note or rest, ones that correspond to the i-th note or rest from the end of the note string, and so on. Furthermore, the note sequence may be modified based on the received note sequence, and the correspondence of the notes may be determined between the transformed note sequences using the above correspondence. An example of modification of the note sequence is to generate a plurality of changed note sequences from the original note sequence by inserting an extra note or deleting the note at each note position.

The note coincidence calculation means 3 receives the two note strings from the input means 1 and the note correspondence calculation means 2 to compare the two note strings with each other. A note-by-note comparison between two note strings is performed to determine the degree of coincidence between notes. The number of matching notes can be used as the note matching degree. As the number of matching notes, for example, the number of notes with the same pitch value or pitch value, the number of notes with both the pitch value and the pitch value at the same time, and the pitch difference from the immediately preceding note You can use the number of notes you have. When the note string includes rests, a linear sum of the number of matching notes and the number of matching rests can be used for the degree of note matching. As the number of matching rests,
You can use the number of rests in the note sequence that match each other, or the number of rests that match in length. Furthermore, as for the note coincidence, a score may be determined according to the degree of coincidence.
That is, in the case of a perfect match, in the case of a slight shift, and in other cases, the score is determined in advance.

The similarity correction amount calculation means 4 receives from the input means 1 the two note sequences and the note correspondence calculation means 2 receives the correspondence relation of the notes to be compared between the two note sequences, and based on the correspondence relation, A similarity correction amount of the similarity between the note strings according to the continuous similar part of the two note strings is calculated. As the similarity correction amount, for example, the number of notes that are completely and continuously matched can be used. If there are rests in the note sequence, it is also possible to use a number in which the notes and rests are completely matched in succession.

The similarity calculating means 5 receives the note matching degree from the note matching degree calculating means 3, and calculates the similarity correction amount calculating means 4
Receives the similarity correction amount from. Then, the note string similarity is calculated from the note matching degree and the similarity correction amount. The similarity between the note strings is, for example, a linear sum of the note coincidence and the similarity correction amount. Further, when the note correspondence calculation means obtains a plurality of correspondences, the correspondence similarity based on the correspondence of each note is obtained, and the maximum value of the correspondence similarity for all correspondences is set as the note string similarity. You can also

The output means 6 receives the similarity between note sequences from the similarity calculation means 5 and outputs it to another device such as a display device or a printer device.

Next, the operation of this embodiment will be described with reference to FIGS.

Two piggyback trains to be compared are input to the input means 1 (step A1). Next, the note correspondence calculation means 2 obtains the correspondence between notes from the input note string (step A2). Next, the note coincidence degree calculation means 3 obtains the degree of coincidence of each note from the correspondence between the inputted note sequence and the note (step A3). Next, the similarity correction amount calculation means 4 calculates the similarity correction amount according to the degree of similarity between consecutive notes based on the correspondence between the inputted note string and the notes (step A). Next, the similarity calculation means 5 obtains the corresponding similarity based on the correspondence of the notes from the note matching degree and the similarity correction amount (step A5). Furthermore, note correspondence calculation means 2
If there is another correspondence relationship obtained by, the correspondence similarity is calculated for other note correspondence relationships (step A).
6). Then, the similarity between the note strings is calculated from the correspondence similarities based on all the note correspondences (step A7). Finally, the output means 6 outputs the similarity between note sequences (step A8).

In the above description of the operation, the calculation is performed in the order of step A3 and step A4, but the calculation may be performed in the order of step A4 and step A3, and step A3 and step A4 are performed simultaneously. You may do it.

Next, referring to FIGS. 1 and 3A,
In the present embodiment, as the note correspondence, the one that compares the notes sequentially from the beginning of the note string is used, the pitch difference from the immediately preceding note is used as the note coincidence, and the complete match is continuously obtained as the similarity correction amount. The case of using the number of notes will be described using a specific example.

The note sequence 1 and the note sequence 2 are given from the input means 1. First, the note correspondence calculation means 2 obtains a correspondence by comparing the i-th notes from the beginning. The note coincidence calculation means 3 compares the pitch difference from the immediately preceding note between the note sequence 1 and the note sequence 2, obtains the value 3 which is the number of notes in which the notes match, and uses this value as the note coincidence. To do. The similarity correction amount calculation means 4 obtains the value 2, which is the number of notes that are in continuous agreement, and sets this value as the similarity correction amount. In the similarity calculation means 5, the note coincidence degree and the similarity correction amount are summed to obtain a value 5, which is output to the output means 6 as the note string similarity degree.

Further, in FIG. 3B, when the note sequence 1 and the note sequence 3 are given to the input means 1, the note coincidence is 3 and the similarity correction amount is calculated by the same calculation as above. It becomes 3. Therefore, the note string similarity is 6. In this way, the note sequence 2 and the note sequence 3 can obtain the result that the note sequence 3 is more similar to the note sequence 1.

In the case of calculating a plurality of correspondences by the note correspondence calculating means 2, it is possible to obtain the correspondence relationship and the correspondence similarity and the similarity between note strings while deforming the note string. In such a case, the degree of similarity between note strings can be obtained using dynamic programming.

FIG. 4 is a block diagram showing another embodiment of the first invention. In addition to the embodiment shown in FIG. 1, the present embodiment is provided with a note string deformation deduction means 11 for obtaining a note string deformation deduction point due to deformation of a note string. Further, the similarity calculation means 5 calculates not only the note coincidence and the similarity correction amount but also the note sequence deformation deduction point to calculate the similarity.

The note string deformation deduction means 11 receives the note correspondence from the note correspondence means 2. Find out how much the correspondence of the notes that has changed due to the transformation of the note sequence has changed compared to the correspondence of the notes that have the original note sequence, and calculate the deduction point for the note sequence transformation, which is the deduction point due to the transformation of the note sequence. . As the note string deformation deduction point, for example, a value obtained by previously multiplying the number of times of note insertion and deletion by a negative number can be used.

The similarity calculating means 5 receives the note matching degree from the note matching degree calculating means 3 and calculates the similarity correction amount calculating means 4
The degree of similarity correction is received from the note sequence transformation point deduction means 11
Receive deduction points from note sequence transformation. Then, the inter-note string similarity is calculated from the note coincidence, the similarity correction amount, and the note string deformation deduction point. The degree of similarity between note strings is, for example, the sum of the degree of note matching, the degree of similarity correction, and the note string deformation deduction point. Further, when the note correspondence calculation means obtains a plurality of correspondences, the correspondence similarity based on the correspondence of each note is obtained, and the maximum value of the correspondence similarities for all the correspondences is set as the note string similarity. You can also

Next, the operation of this embodiment will be described with reference to FIGS. 4 and 5.

Operation of the input means 1, the note correspondence calculation means 2, the note coincidence calculation means 3 and the similarity correction amount calculation means 4 (FIG. 5, step A1 to step A4 and step A)
6 to step A8) are the same as those in the first embodiment of the present invention 1, the description thereof will be omitted. In the first embodiment, the calculation of the correspondence similarity based on the correspondence between the notes is obtained only by the note coincidence and the similarity correction amount. In the present embodiment, the note string deformation deduction calculation means 11 obtains a note string deformation deduction point associated with the note string deformation (step B1). Similarity calculation means 5
Then, from the note matching degree, the similarity correction amount and the note string deformation deduction point,
The correspondence similarity based on the correspondence of the notes is obtained (step 5).

In the above description of the operation, the calculation is performed in the order of step A3, step A4 and step B1. The calculation may be performed sequentially from any step, and the calculation may be performed simultaneously. Furthermore, when calculating a plurality of correspondences with the note correspondence calculation means 2,
It is also conceivable to obtain the correspondence relationship and the similarity degree between the note sequences while performing the transformation of the note sequence. In such a case, the degree of similarity between note strings can be obtained using dynamic programming.

Next, an embodiment of the second invention will be described with reference to the drawings. In the second invention, since the similarity correction amount calculating means 4 of the first invention is performed by another means, this part will be described. FIG. 6 is a block diagram for implementing the similarity calculation means 4 of the second invention.

The input means 41 receives two note strings A and B.

The plural note pitch difference calculating means 42 receives the note sequence A and the note sequence B from the input means 41. For example, in the case of obtaining the pitch difference between each note and the n-preceding note, when the number of both notes is I, the plural-preceding pitch difference is calculated for the i-th note position where n <i <I. Ask. The pre-pitch differences D _A (i) and D _B (i) can be obtained as follows, for example.

D _A (i) = A (i) −A (i−n) n <i <I D _B (i) = B (i) −B (i−n) n <i <I plural notes before The pitch difference matching degree calculation means 43 receives the pitch difference sequences D _A and D _B from the plural-note pitch difference calculation means 42 from the plural-note preceding notes of the note strings A and B, and receives the plural-note preceding pitches. The degree of difference matching is obtained, and the degree of similarity correction is obtained from the degree of previous pitch difference agreement. A method of obtaining the degree-of-pitch difference before plural pitches is, for example, n <
_A method may be used in which if D _A (i) and D _B (i) match i for i <I, then a predetermined value x _n , and if not, y _n . For the similarity correction amount, for example, the sum of the plural pitches of the pitch differences before each note obtained for each note for i <2 <i <I can be used.

The output means 44 receives the similarity correction amount from the plural-pitch preceding pitch difference matching degree calculation means 43 and holds it.

In the above description, n is fixed, the pitch difference between each note and the sound of the n-note is calculated, and the pitch difference before n-note is compared to obtain the degree-of-pitch matching before n-note. Although the description has been given with the similarity correction amount, the value of n is changed such that n = 2, ... The sum of the pre-n-note pitch difference matching degrees may be used as the similarity correction amount. In order to obtain the same similarity correction amount, first, the n-note pitch difference for each note is found for n = 2, ..., N, and the obtained plural-note pitch difference is used. The degree of coincidence may be obtained, and the degree of similarity correction may be obtained from the degree of coincidence of pitch differences between plural notes.

Next, referring to FIGS. 1, 4 and 7 (A), in the present embodiment, the one in which the notes are compared in order from the beginning of the note string is used as the note correspondence, and the note matching degree is the note. A specific example will be used to describe the case where the number of matching pitch differences with the preceding sound is used and the number of matching two preceding pitch differences is used as the similarity correction amount.

The note sequence 1 and the note sequence 2 are given from the input means 1. First, the note correspondence calculation means 2 obtains a correspondence by comparing the i-th notes from the beginning. The note coincidence calculation means 3 compares the pitch difference from the immediately preceding note between the note sequence 1 and the note sequence 2, obtains the value 3 which is the number of notes in which the notes match, and uses this value as the note coincidence. To do. In the similarity correction amount calculation means 4, first, the plural-pitch preceding pitch difference calculation means 2
In 2, the pitch difference between each note and the note two notes before is determined. Next, the corresponding notes are compared with each other in terms of the degree of coincidence of the two-note preceding note, and the number of coincident notes 1 is obtained. In the similarity calculation means 5, the note coincidence and the similarity correction amount are summed to obtain a value 4, which is output to the output means 6 as the similarity between note strings.

Further, in FIG. 7B, when the note sequence 1 and the note sequence 3 are given to the input means 1, the note coincidence is 3 and the similarity correction amount is 2 by the same calculation as the above.
A value of 2 is obtained from the pre-pitch difference in matching. Therefore, the degree of similarity between note strings is 5. In this way, note sequence 2 and note sequence 3
Can obtain the result that note sequence 3 is more similar to note sequence 1.

In the above embodiment, in order to obtain the similarity correction amount, first, the pitch difference before a plurality of pitches is calculated, then the pitch difference matching degree before the plurality of pitches is calculated, and finally the similarity correction amount is calculated. However, the order of calculation is not important. That is, the plural-pitch difference calculation unit 42 obtains the plural-pitch difference of the i-th note, and the plural-pitch difference matching degree calculation unit 4
In step 3, the calculation of the degree-of-preceding pitch difference matching between the i-th notes is repeated for 2 <i <I, and from this result, the degree-of-preceding pitch difference matching calculation means 43 calculates the similarity correction amount. The same result can be obtained by calculation.

[0043]

As described above, according to the first aspect of the present invention, the similarity can be changed by the continuous similarity portion when comparing the note sequences. According to the second invention, in order to compare a note before a plurality of notes and a pitch difference between note sequences, a high similarity correction value can be obtained by comparison between note sequences having continuous similar portions. Therefore, by using the present invention, it is possible to perform an appropriate similarity calculation according to the degree of continuous similarity, as compared with the conventional method that only obtains the ratio of similar portions to the whole.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of the first invention.

FIG. 2 is a flow chart for explaining a first embodiment of the first invention.

FIG. 3 is an explanatory diagram showing an operation of the first exemplary embodiment of the first invention.

FIG. 4 is a block diagram showing a second embodiment of the first invention.

FIG. 5 is a flow chart for explaining a second embodiment of the first invention.

FIG. 6 is a block diagram showing an embodiment of the second invention.

FIG. 7 is an explanatory diagram showing an operation of one embodiment of the second invention.

FIG. 8 is a block diagram showing a conventional example.

FIG. 9 is an explanatory diagram showing an operation of a conventional example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 input means 2 note correspondence calculation means 3 note coincidence calculation means 4 similarity correction amount calculation means 5 similarity calculation means 6 output means 11 note string deformation deduction calculation means 41 input means 42 plural-note pitch difference calculation means 43 plural Pre-pitch difference calculation means 44 Output means

Claims (2)

[Claims] 1. Between note sequences represented as a sequence of notes
In order to perform comparison and matching, for each note included in the note string,
In a note string similarity calculation device that obtains a note string similarity using a comparison / matching result, a correspondence of notes to be compared and matched between the note strings is obtained for the entire note string.
Note correspondence calculation means and the note interval based on all of the correspondence
Based on the result of comparison and matching in the
A note matching degree calculating means for obtaining a note matching degree indicating the degree of matching of the following notes, and a case where the notes consecutively match between the note strings based on the correspondence.
The similarity correction amount according to the degree of continuous similarity
An inter-note sequence similarity calculation device comprising: a similarity correction amount calculating means; and a similarity calculating means for calculating a similarity between note sequences from the note matching degree and the similarity correction amount. 2. A plurality of pitch differences, which is a difference between a pitch value of a note included in a note string based on the correspondence and a pitch value of a note preceding a plurality of notes of the note, by the similarity correction amount calculation means. 2. The musical note sequence according to claim 1, further comprising a plural-note pitch difference calculating means for obtaining a value and a plural-note pitch difference matching degree calculating means for obtaining a matching degree of the preceding pitch difference values. Inter-similarity calculator.