JP3612272B2 - Music information search device, music information search method, and computer-readable recording medium storing music information search program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention is a feature amount of music information.ForMusic information search device that searches music informationAnd a method thereof, and a computer-readable recording medium storing a music information search program used for realizing the music information search methodRelated.
[0002]
[Prior art]
Similarity search is an indispensable technology for building multimedia databases.
[0003]
For example, when searching for an image with an apple, there is rarely a need to search for an image with an apple that is exactly the same as the apple picture or photo given as a search key. Most of them are requests that simply want an image showing an apple. In such a case, a similar search technique such as “Find an image similar to the apple image given as a search key” is required. .
[0004]
In order to deal with such new media for content search, research on techniques for searching song titles and singers from a music information database using humming and melody is popular.
[0005]
Against this background, the present inventors have also extracted a feature quantity related to sound transition and sound distribution from music and used it to search a music information database (Naoko Kosugi, Yuichi Nishihara, Seichiki Kon'ya, Masashi Yamamura, and Kazuhiko Kusima: Music Retrieval by Humming-Veimetic Retrieval
[0006]
In the music search system proposed by the present inventors, a person's singing is accepted as a search key, so that a person who only remembers only a part of the melody of the song searches for the title of the song, a singer / performer, etc. Offering to be able to.
[0007]
When searching for information related to a song by a person's singing, a person does not necessarily sing rust or music. Therefore, the music search system proposed by the present inventors can divide music data (melody data) finely and increase the redundancy of the music data to be held so that a person can search any part of a song. Has been realized.
[0008]
In realizing the configuration, in the music search system proposed by the present inventors, as shown in FIG. 21, music pieces are cut out by dividing the music data into predetermined intervals, and each piece of music is cut out. The feature amount (feature amount vector) starting from the beginning of the music piece is calculated.
[0009]
Then, a music information database is constructed using the feature quantity calculated in this way, and stored in the music data and music information database of the person's song using the feature quantity thus calculated. The configuration is such that the degree of similarity with music data is evaluated.
[0010]
As described above, the conventional technology adopts a configuration in which the feature amount (feature amount vector) is calculated from the entire music piece starting from the beginning of the music piece.
[0011]
Note that the feature amount (feature amount vector) calculated by this conventional technique may be hereinafter referred to as “Entire Feature Vector” in consideration of being calculated from the entire music piece.
[0012]
[Problems to be solved by the invention]
However, according to such a conventional technique, there is a problem that the search accuracy is remarkably deteriorated when the singing part of the searcher does not coincide with the head of the divided music data (melody data).
[0013]
That is, in the prior art, since the feature amount is calculated from the whole music piece starting from the beginning of the music piece, the singing part of the searcher and the beginning of the divided music data (melody data) If the and do not match, the search accuracy is significantly deteriorated.
[0014]
In order to solve this problem, it is conceivable to use a method of dividing music data more finely. However, if this method is used, a new problem that the amount of data increases explosively appears. come.
[0015]
The present invention has been made in view of such circumstances, and it is possible to improve the search accuracy when a system for searching for information related to a song by human singing is constructed.NewProviding music information retrieval technologyServingObjective.
[0023]
[Means for Solving the Problems]
ThisIn order to achieve the above object, the music information retrieval apparatus of the present invention provides a plurality of music cut out from input music information.For each piece, a feature defined as a temporal position on the piece of music that satisfies a given condition.Generating means for detecting feature points and generating feature quantities starting from those feature points; and music to be searchedSpecial features of songsCollection amount (features to be generated by the generation means) MemorySteppedThe stored data in the storage meansGeneration of generation means by searchingdidSearch means for searching for a plurality of music pieces having similar feature quantities for each feature quantity, and search by the search meansdidA specifying means for specifying the music indicated by the input music information from the music is provided.
[0024]
Here, the generation means may detect a feature point of one piece of music cut out from the input music information and generate a feature quantity starting from the feature point. The search means searches for one or a plurality of music pieces having a feature quantity similar to the generated feature quantity and outputs it as a search result.
[0025]
When configured in this way, conversion means for converting the input music information into a prescribed note form (for example, a prescribed note form such as an eighth note capable of improving the search accuracy verified by experiments) In this case, the generation unit generates a feature amount using a music piece cut out from the music information converted by the conversion unit as a processing target.
In addition, the generation unit may detect a feature point as a section used for detecting a feature point using a section having a specified length located at a specified temporal position of the music piece. At this time, the generation unit may detect the feature point. By detecting a sound position that satisfies a condition regarding the pitch or length of a sound imposed on the sound, a feature point included in the section may be detected.
The generation unit generates a feature amount from a music piece portion having a specified length starting from the feature point, or generates a feature amount from a music piece portion starting from the feature point and ending at the end of the music piece.
In consideration of the fact that the feature quantity (feature quantity vector) calculated by the music information retrieval apparatus of the present invention is calculated from a part of a music piece, “ Partial Feature Vector May be called.
In the music information retrieval apparatus of the present invention configured as described above, the generation means detects each feature point of a plurality of music pieces cut out from the input music information, and uses those feature points as starting points. Generate feature values.
[0026]
In response to the generation of this feature quantity, search meansThe stored data of the storage meansGenerating means by searchingdidFor each feature quantity, a plurality of music pieces having similar feature quantities are searched, and in response to this, the specifying means performs, for example, an OR operation on the similarity degree of the searched music pieces, thereby obtaining the input music information. Identify the song you are pointing to.
[0027]
In order to improve the search accuracy when adopting this configuration, the generation means converts the feature quantity of the generated time-series data format into another expression format such as a histogram format, for example. A plurality of feature quantities may be generated, and in response to this, the database means may manage a plurality of feature quantities having different representation formats.
[0028]
In this way, the music information retrieval apparatus according to the present invention does not adopt a configuration in which the feature amount is calculated from the entire music piece as in the prior art, but as shown in FIG. By adopting a configuration for detecting included feature points, processing is performed so as to calculate a feature amount that can be called a partial feature amount of a musical piece starting from the feature point.
In the music information retrieval apparatus of the present invention,As shown in Fig. 2, it has the music to be searched.Calculated in such a calculation form.A music information database that is managed in correspondence with the feature quantityAccording to such a calculation form,Features of the input music informationRawThe music obtained by dividing the input music information is processed by searching the music information database using it as a search key and searching for the music indicated by the input music information. Even when the heads of the pieces and the divided pieces of music stored in the music information database do not match, high search accuracy can be realized.
[0029]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the present invention will be described in detail according to embodiments.
[0030]
FIG. 3 shows an example of an embodiment of the music information retrieval apparatus 1 comprising the present invention.
[0031]
The music information search apparatus 1 of the present invention performs a process of searching for information (song name, etc.) relating to a person's hummed music. In order to realize this search process, a music information database 10 and a music recording unit 11 And a noise removal unit 12, a feature amount generation unit 13, and a search unit 14.
[0032]
Here, the music recording unit 11, the noise removal unit 12, the feature amount generation unit 13, and the search unit 14 included in the music information search device 1 of the present invention are specifically realized by programs, and these programs are It can be stored in an appropriate recording medium such as a semiconductor memory readable by a computer.
[0033]
The music information database 10 manages the music information (song name / melody information) of the music to be searched, the feature amount of the music, and the attribute information (information of singer / player, etc.) of the music. The feature quantity of the music managed in the music information database 10 is generated and registered by the feature quantity generation unit 13.
[0034]
The music recording section 11 is composed of, for example, commercially available music recording software, and creates input Hamming score information.
[0035]
The musical score information created by the transcription unit 11 is created in MIDI format, and is composed of information relating to the relative time constituting the musical score information and information relating to the absolute time constituting the performance time of the musical score information. However, in order to normalize the time with the feature quantity (described in relative time) managed by the music information database 10, the transcription unit 11 is a part not related to the absolute time. A certain musical score information part is processed so as to be passed to the noise removing unit 12.
[0036]
The noise removing unit 12 receives the music information (score information) passed from the transcription unit 11 and continues a sound that is off the center melody line or a silent section of a certain length or longer in a section where recording starts. When this is done, a process is performed to remove noise input to humming to be input, such as by deleting a sound section before the silent section. For example, when humming as shown in FIG. 4 is input, processing is performed so that the first, fourth, and fourteenth sounds are removed as noise.
[0037]
The feature value generation unit 13 divides the music information passed from the noise removal unit 12 into music pieces, and a music feature value for calculating the feature values (melody feature values) of the divided music pieces. A calculation unit 131, a feature point selection section limiting unit 132 that limits a section for selecting feature points of a musical piece from the calculated feature quantity (melody feature quantity), and feature points from the limited sections. A feature amount determination unit 133 to be determined and a feature amount deriving unit 134 to derive a feature amount starting from the determined feature point are generated to generate a feature amount of the input hamming and search it 14 is executed.
[0038]
Here, the music feature amount calculated by the music feature amount calculation unit 131 is a feature amount associated with the melody on a one-to-one basis, whereas the feature amount derived by the feature amount deriving unit 134 is characteristic of the present invention. It is a characteristic amount. Hereinafter, for convenience of explanation, the feature amount derived by the feature amount deriving unit 134 may be referred to as a “feature point starting music partial feature vector”.
[0039]
The search unit 14 searches the music information database 10 using the feature value of Hamming passed from the feature value generation unit 13 as a search key, thereby searching for the input Hamming song name and the like.
[0040]
FIG. 5 illustrates an example of a processing flow executed by the feature value generation unit 13. Next, processing executed by the feature value generation unit 13 will be described according to this processing flow.
[0041]
When the humming music information is passed from the noise removing unit 12, the feature amount generating unit 13 is first started in step 1 as shown in the processing flow of FIG. Enter the following.
[0042]
Subsequently, in step 2, the inputted music information is divided into music pieces by cutting out a piece of music having a length of 16 beats from the inputted music information, for example, while shifting by 4 beats. This division process corresponds to the process of the music information division unit 130.
[0043]
Subsequently, in step 3, the feature value of the melody of each music piece is calculated. For example, when the feature value of the melody of a musical piece is calculated using a MIDI code indicating a pitch as shown in FIG. 6 as a specific example, “64, 64, 65, 65, 67” will be described. , 67, 65, 65 ", or a melodic feature value of" -1, -1, 0, 0, 2, 2, 0, 0 "with 65 as a base tone. To do. This calculation process corresponds to the process of the music feature amount calculation unit 131.
[0044]
Subsequently, when it is determined in step 4 whether or not all the generated music pieces have been processed and it is determined that all the music pieces have not been processed, the process proceeds to step 6 where unprocessed music pieces are processed. Select one.
[0045]
Subsequently, in step 7, the section of the central portion of the selected music piece is limited (extracted) as a section for selecting feature points. For example, the central four beats of a musical piece having a length of 16 beats are limited. This limitation process corresponds to the process of the feature point selection section limitation unit 132.
[0046]
Subsequently, in step 8, the highest sound in the limited section is determined (selected) as a feature point. This determination process corresponds to the process of the feature point determination unit 133.
[0047]
Subsequently, in step 9, the feature point from the determined feature point to the last beat of the musical piece is derived (generated) as the characteristic amount (feature point starting music partial feature vector) of the musical piece, and then the next musical piece is processed. Return to step 4 as much as possible. This derivation process corresponds to the process of the feature quantity derivation unit 134.
[0048]
When it is determined in step 4 that all the generated music pieces have been processed, the process proceeds to step 5 where the generated feature amount (obtained for each music piece) is passed to the search unit 14 and the process is terminated. .
[0049]
In this way, the feature value generation unit 13 cuts out a piece of music having a length of 16 beats from the input music information while shifting it by 4 beats to obtain the feature value of the melody when the processing flow of FIG. 5 is followed. When calculated, as shown in FIG. 7, first, the central four beats of the musical piece are limited, then the highest sound among the four beats is determined as a feature point, and then the feature From the point to the last beat, processing is performed so as to derive the feature value (feature point starting music partial feature vector) of the music piece.
[0050]
FIG. 8 illustrates another embodiment of the processing flow executed by the feature quantity generation unit 13.
[0051]
In the case of following this processing flow, when the feature quantity generation unit 13 is started by receiving the humming music information from the noise removal unit 12, first, the feature information generation unit 13 inputs the music information in step 1 and continues. In step 2, music pieces are cut out from the input music information, and in step 3 the melodic feature value of each music piece is calculated.
[0052]
Subsequently, when it is determined in step 4 whether or not all the generated music pieces have been processed and it is determined that all the music pieces have not been processed, the process proceeds to step 6 where unprocessed music pieces are processed. Select one.
[0053]
Subsequently, in step 7, the first 4-beat section of the selected musical piece is limited (extracted) as a section for selecting feature points, and in step 8, the longest sound in the first 8 beats is selected. In addition, a sound whose sound begins in the limited section (first 4 beats) is determined (selected) as a feature point, and in step 9, from the determined feature point to 12 beats, the feature value of the music piece ( After being derived (generated) as a feature point starting music partial feature vector, the process returns to step 4 to process the next music piece.
[0054]
When it is determined in step 4 that all the generated music pieces have been processed, the process proceeds to step 5 where the generated feature amount (obtained for each music piece) is passed to the search unit 14 and the process is terminated. .
[0055]
In this way, the feature value generation unit 13 cuts out a piece of music having a length of 16 beats from the input music information while shifting it by 4 beats to obtain the feature value of the melody when the processing flow of FIG. 8 is followed. When calculated, as shown in FIG. 9, first the first 4 beats of the piece of music are limited, followed by the longest sound within the first 8 beats, and the ringing is at the first 4 beats. The sound is determined as a feature point, and then, 12 beats from the feature point are processed so as to be derived as the feature value of the music piece (feature point starting music partial feature vector).
[0056]
As described above, the feature quantity generation unit 13 is not configured to calculate the feature quantity from the entire music piece as in the prior art, but is configured to detect the feature points included in the music piece. Then, processing is performed so as to derive a feature amount (feature point starting music partial feature vector) starting from the feature point.
[0057]
That is, in the case of the music information shown in FIG. 10, with respect to a music piece cut out from the 18th note, in the prior art, the entire music piece is started from the 18th note located at the beginning of the music piece. The feature value generation unit 13 generates, for example, a “Partial Feature Vector” from “Partial Feature Vector” from a part of the music piece, for example, starting from the 19th note of the music piece. A feature amount to be expressed as “Feature Vector” is generated.
[0058]
Although not shown in FIG. 3, various processes are prepared for the processes executed by the feature point selection section limiting unit 132, the feature point determination unit 133, and the feature quantity deriving unit 134. It is desirable to be able to select an optimum one that realizes the improvement of the above.
[0059]
In order to respond to such a request, as shown in FIG. 11 (a), a plurality of rules for limiting feature point selection sections according to different algorithms are prepared, and a rule set by the user is called out of them. As shown in FIG. 11B, a plurality of rules for determining feature points according to different algorithms are prepared, and a rule set by the user is called out from the rules. As shown in FIG. 11 (c), a plurality of rules for deriving feature quantities (feature point starting music partial feature vectors) according to different algorithms are prepared, It is preferable to use a configuration in which the feature amount is derived by calling a rule set by the user.
[0060]
Also, the feature quantity generation unit 13 is a feature point starting music partial feature as shown in FIG. 12B defined by MIDI time-series data indicating the pitch of the sound as shown in FIGS. When generating a vector, a feature point starting music partial feature vector as shown in FIG. 12B defined by the histogram of the MIDI code may be generated.
[0061]
As will be described later, if the music information database 10 is searched using a plurality of feature point starting music partial feature vectors having such different expression formats, the search accuracy can be improved. As shown in FIG. 12, when generating a feature point starting music partial feature vector defined by time series data, another feature point starting music partial feature vector having a different expression form is then generated. It may be processed.
[0062]
13 and 14 show an embodiment of a processing flow executed by the search unit 14. Here, FIG. 13 illustrates processing that the search unit 14 executes when the feature amount generation unit 13 generates one type of feature amount. FIG. 14 is different in the feature amount generation unit 13. The process which the search part 14 will perform when producing | generating the some feature-value with an expression form is illustrated.
[0063]
Next, search processing executed using the feature amount (feature point starting music partial feature vector) generated by the feature amount generation unit 13 in accordance with these processing flows will be described.
[0064]
In the case of following the processing flow of FIG. 13, when the search unit 14 is activated by passing the feature values of a plurality of music pieces cut out from humming from the feature value generation unit 13, first, at step 1, , One type of feature value of each music piece is received from the feature value generation unit 13.
[0065]
Subsequently, in step 2, it is determined whether or not all music pieces have been selected, and when it is determined that unprocessed music pieces remain, the process proceeds to step 3 where unprocessed music pieces are selected. Select one.
[0066]
Subsequently, in step 4, the feature quantity of the selected music piece is set as a processing target, and in the subsequent step 5, the music information database 10 is searched to have a feature quantity similar to the feature quantity of the processing target. For example, search for the top 20 songs and return to step 2 to process the next piece of music.
[0067]
On the other hand, when it is determined in step 2 that all pieces of music have been selected, the process proceeds to step 6 where the similarity of the searched music (for example, when evaluating the similarity using distance, the distance value is small). For example, the top 10 songs that are similar to Hamming are specified by performing an OR operation on the value.
[0068]
For example, as shown in FIG. 15, from the feature amount of the music piece 1, music D having a similarity of distance 0.9 and music B having a similarity of distance 1.5 / similarity of distance 1.8 Music A having a similarity of distance 5.8 is searched, and music A having a similarity of distance 0.3 and music B / having a similarity of distance 1.2 from the feature amount of music piece 2 are searched. A song C having a similarity of distance 2.0 and a song D having a similarity of distance 5.9 are searched, and from the feature quantity of the music piece 3, a song B having a similarity of distance 1.0 and a distance 1. When a song C having a similarity of 2 / a song D having a similarity of 1.5 and a song D having a similarity of 1.5 and a distance of 6.0 is searched, the minimum distance of the song A is 0.3. Since the minimum distance of music B is 1.0, the minimum distance of music C is 1.2, and the minimum distance of music D is 0.9, According, A and song most similar to Hamming, then D as music to be similar, B as music to be similar to the next, is to identify the C as music to be similar to the next.
[0069]
Finally, in step 7, the identified top 10 songs, for example, are output as search results, and the process ends.
[0070]
Further, in the case of following the processing flow of FIG. 14, when the search unit 14 is started by passing the feature values of a plurality of music pieces cut out from humming from the feature value generation unit 13, first, the search unit 14 performs steps. 1, a plurality of types of feature values of each music piece are received from the feature value generation unit 13.
[0071]
Subsequently, in step 2, it is determined whether or not all music pieces have been selected, and when it is determined that unprocessed music pieces remain, the process proceeds to step 3 where unprocessed music pieces are selected. Select one.
[0072]
Subsequently, in step 4, it is determined whether or not all the feature values of the selected music piece have been processed, and when it is determined that unprocessed ones are left, the process proceeds to step 5, From the feature quantities of the selected music piece, an unprocessed one is selected as a processing target, and in step 6, the music information database 10 is searched to obtain a feature quantity similar to the processing target feature quantity. Search and return to step 4 to process the next feature.
[0073]
In this way, when it is determined in step 4 that all of the feature values of the selected music piece have been processed by repeating the processing of step 4 to step 6, the process proceeds to step 7. The weighted linear sum of the similarities of the searched feature quantities is calculated, and, for example, the top 20 feature quantities are specified, and the top 20 songs having the feature quantities are searched for. The process then returns to step 2 to process the next piece of music.
[0074]
On the other hand, when it is determined in step 2 that all pieces of music have been selected, the process proceeds to step 8 where the above-mentioned OR operation is performed on the similarity of the searched music pieces, so that, for example, the top 10 similar to Hamming The music is specified, and in step 9 that follows, the specified top 10 songs, for example, are output as search results, and the process is terminated.
[0075]
In this way, when the feature values of a plurality of music pieces cut out from the humming are passed from the feature value generation unit 13, the search unit 14 searches the music information database 10 using the feature values as search keys. For example, the top 10 songs similar to humming are obtained and processed so as to be output.
[0076]
In the processing flow of FIG. 13 and FIG. 14, the feature value generation unit 13 cuts out a plurality of music pieces from the humming, generates the feature value, and passes it to the search unit 14. For example, the feature amount generation unit 13 may generate a feature amount of only one piece of music located in the middle from the hamming and pass the result to the search unit 14.
[0077]
At this time, the search unit 14 executes the processing flow of FIG. 16A instead of the processing flow of FIG. 13, and executes the processing flow of FIG. 16B instead of the processing flow of FIG. become. At this time, the OR operation described in FIG. 15 is not performed.
[0078]
Next, the results of experiments conducted to verify the effectiveness of the present invention will be described.
[0079]
This experiment is performed when the music information database 10 is configured to manage information on about 10,000 songs, and when 25 users hum a total of 186 songs selected by the user, The search results were obtained up to the 10th song title, and the song title was checked against the hummed song.
[0080]
Here, the feature point starting music partial feature vector is limited to the first four beats of the music piece, the highest sound among the four beats is determined as the feature point, and 12.5 beats from the feature point are determined. What was obtained by extracting was used.
[0081]
FIG. 17 compares the experimental results obtained when the feature point starting music partial feature vector obtained by the present invention is used as a search key and the experimental results obtained when “Entire Feature Vector” obtained by the prior art is used as a search key. Show.
[0082]
As can be seen from the experimental results, when the feature point starting music partial feature vector obtained by the present invention is used as a search key, the probability that a hummed song will enter the 10th place is about 75%, whereas according to the conventional technique. When the obtained “Entire Feature Vector” is used as a search key, the probability that a hummed song will enter the 10th place is only about 65%, and thus the effectiveness of the present invention can be verified.
[0083]
FIG. 18 shows a comparison between an experimental result when a search is performed by cutting out a plurality of music pieces from Hamming, and an experimental result when a search is performed by cutting out one music piece located in the middle of Hamming. When the former search is performed, processing is performed so as to obtain a final search result by performing the OR operation described in FIG.
[0084]
As can be seen from the experimental results, it has been verified that the search accuracy can be improved by performing a search by cutting out a plurality of music pieces from Hamming than by performing a search by cutting out a single music piece.
[0085]
FIG. 19 shows an experimental result when a feature point origin music partial feature vector (Partial Tone Transition Feature Vector) is used as a search key, and a feature vector (Tone Distribution Feature) obtained by converting the feature point origin music partial feature vector into a histogram representation format. An experimental result when using Vector) as a search key and an experimental result when using a combination of the two feature vectors as a search key are shown in comparison.
[0086]
As can be seen from the experimental results, if the search key is the feature point starting music partial feature vector and the feature vector obtained by converting the feature point starting music partial feature vector into a histogram representation format, only the feature point starting music partial feature vector is used as the search key. It was verified that the search accuracy could be improved more than the case.
[0087]
FIG. 20 compares the experimental result when the feature value managed by the music information database 10 is described in quarter-notes with the experimental result when described in quarter-notes. Show. When the humming note form is different from the note form of the music information database, the feature value generating unit 13 converts the humming note form into the note form of the music information database, and then performs the above-described processing. .
[0088]
As can be seen from the experimental results, it has been verified that the search accuracy can be improved by realizing the present invention in the form of an eighth note than in the form of a quarter note.
[0089]
From now on, although not explained in the above-described embodiment, the feature quantity managed by the music information database 10 is constructed in a note form that can improve the search accuracy like an eighth note, and the feature quantity generation unit 13 It is preferable to have a function of converting the humming note form into the note form of the music information database 10.
[0090]
【The invention's effect】
As described above, the present invention adopts a configuration for detecting feature points included in a music piece, instead of adopting a configuration for calculating feature amounts from the whole music piece as in the prior art. Since the processing is performed so as to calculate the feature quantity that can be called the partial feature quantity of the music piece starting from the feature point, the music information database is searched for something similar to the input music information.PlaceTogetherInEven when the music piece obtained by dividing the music information that has been applied does not coincide with the beginning of the divided music piece stored in the music information database, high search accuracy can be realized.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram of the present invention.
FIG. 2 is an explanatory diagram of the present invention.
FIG. 3 is an example of an embodiment of the present invention.
FIG. 4 is an example of a score for explaining noise removal processing;
FIG. 5 is an example of a processing flow executed by a feature amount generation unit.
FIG. 6 is an explanatory diagram of a calculation process of a feature value of a music piece melody;
FIG. 7 is an explanatory diagram of processing executed by a feature amount generation unit.
FIG. 8 is another embodiment of the processing flow executed by the feature quantity generation unit.
FIG. 9 is an explanatory diagram of processing executed by a feature amount generation unit.
FIG. 10 is an example of a score for explaining the present invention.
FIG. 11 is an explanatory diagram of a configuration example of a feature amount generation unit.
FIG. 12 is an explanatory diagram of a feature amount generated by a feature amount generation unit;
FIG. 13 is an example of a processing flow executed by a search unit.
FIG. 14 is an example of a processing flow executed by a search unit.
FIG. 15 is an explanatory diagram of an OR operation process for obtaining a search result.
FIG. 16 is another example of a processing flow executed by the feature quantity generation unit.
FIG. 17 is an explanatory diagram of the results of an experiment performed to verify the effectiveness of the present invention.
FIG. 18 is an explanatory diagram of the results of an experiment performed to verify the effectiveness of the present invention.
FIG. 19 is an explanatory diagram of the results of an experiment performed to verify the effectiveness of the present invention.
FIG. 20 is an explanatory diagram of the results of an experiment performed to verify the effectiveness of the present invention.
FIG. 21 is an explanatory diagram of a prior art.
[Explanation of symbols]
1 Music information retrieval device
10 Music information database
11 Transcription
12 Noise removal unit
13 Feature generator
14 Search part
130 Music Information Division
131 Music feature amount calculation unit
132 Feature point selection section limited section
133 Feature point determination unit
134 Feature Deriving Unit

Claims (13)

In a music information search device for searching for a song name indicated by input music information and / or attribute information of the song,
For one music piece to be cut out from the input music information, by detecting a feature point which is defined as the time position of the piece satisfying a predetermined condition music, it generates a characteristic quantity which starts its feature point generation Means,
A storage means to store the characteristic quantity having the easy songs to be searched,
A search unit for searching for one or a plurality of music pieces having the feature amount similar to the feature amount generated by the generation unit by searching stored data in the storage unit ;
A featured music information retrieval device. In a music information search device for searching for a song name indicated by input music information and / or attribute information of the song,
For each of a plurality of music pieces to be cut out from the input music information, by detecting a feature point which is defined as the time position of the piece satisfying a predetermined condition music, a feature quantity which starts their characteristic points Generating means for generating;
A storage means to store the characteristic quantity having the easy songs to be searched,
Retrieval means for retrieving a plurality of pieces of music having the similar feature quantities for each of the feature quantities generated by the generation means by searching stored data in the storage means;
Including specifying means for specifying the music indicated by the input music information from the music searched by the search means,
A featured music information retrieval device. The music information search device according to claim 2 ,
The specifying means specifies the music indicated by the input music information by ORing the similarity of the music searched by the search means,
A featured music information retrieval device. The music information search apparatus according to any one of claims 1 to 3,
A conversion means for converting the input music information into a prescribed note form;
The generation unit generates a feature amount using a music piece cut out from the music information converted by the conversion unit as a processing target.
A featured music information retrieval device. The music information search device according to any one of claims 1 to 3,
The generating means detects a feature point as a section used for detecting a feature point using a section having a specified length located at a specified time position of the music piece.
A featured music information retrieval device. The music information retrieval apparatus according to claim 5, wherein
The generating means detects a feature point included in the section by detecting a sound position that satisfies a condition for a pitch or length of a sound imposed on the feature point.
A featured music information retrieval device. The music information search apparatus according to any one of claims 1 to 3,
The generation means generates a feature amount from a music piece portion having a specified length starting from the feature point.
A featured music information retrieval device. The music information search device according to any one of claims 1 to 3,
The generating means generates a feature value from a music piece portion starting from a feature point and ending at a music piece end point.
A featured music information retrieval device. The music information search device according to any one of claims 1 to 8 ,
It said generating means, to convert the feature amount the generated into another representation, that in addition to the feature quantity, generates a feature amount other than it,
A featured music information retrieval device. In a music information search method for searching for a song name indicated by input music information and / or attribute information of the song,
A process of detecting a feature point defined as a temporal position on a music piece satisfying a predetermined condition for one piece of music cut out from input music information, and generating a feature amount starting from the feature point; ,
A process of searching for one or a plurality of music pieces having the feature amount similar to the generated feature amount by searching stored data in a storage unit that stores the feature amount of the song to be searched. To prepare,
Characteristic music information retrieval method. In a music information search method for searching for a song name indicated by input music information and / or attribute information of the song,
For each of a plurality of pieces of music cut out from the input music information, feature points defined as temporal positions on the piece of music satisfying a predetermined condition are detected, and feature amounts starting from those feature points are generated. The process of
A process of searching for a plurality of pieces of music having the similar feature amount for each of the generated feature amounts by searching stored data in a storage unit that stores the feature amount of the music to be searched;
A process of identifying the music indicated by the input music information from the searched music,
Characteristic music information retrieval method. A computer-readable recording medium storing a music information search program for searching for a song name indicated by input music information and / or attribute information of the song,
Processing for detecting a feature point defined as a temporal position on a piece of music that satisfies a predetermined condition for one piece of music cut out from the input music information, and generating a feature amount starting from the feature point; ,
A process of searching for one or a plurality of music pieces having the feature amount similar to the generated feature amount by searching stored data of a storage unit that stores the feature amount of the song to be searched. Recording a music information search program to be executed by a computer
A computer-readable recording medium on which a music information search program is recorded. A computer-readable recording medium storing a music information search program for searching for a song name indicated by input music information and / or attribute information of the song,
For each of a plurality of pieces of music cut out from the input music information, feature points defined as temporal positions on the piece of music satisfying a predetermined condition are detected, and feature amounts starting from those feature points are generated. Processing to
A process for searching a plurality of music pieces having the above-mentioned feature amount similar to each of the generated feature amounts by searching stored data of a storage unit that stores the feature amount of the music to be searched;
Recording a music information search program that causes a computer to execute processing for specifying a music pointed to by the input music information from the searched music pieces,
A computer-readable recording medium on which a music information search program is recorded.

Images