JP2000347659A - Music retrieving device and method, and storage medium recorded with music retrieving program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely and quickly retrieve music intended by a retiever out of a database, by inputting freely a portion or the whole of the music without making the retriever be aware of a phrase, a melody, a tempo and the like. SOLUTION: A tempo of input music or musical information converted from the input music is normalized by a tempo normalizing means 2 to be stored in a database by a music accumulating means 4, in database construction. The tempo of the music or the musical information converted from the music, which is a retrieving key, is normalized by the tempo normalizing means 2, similarity retrival is carried out by a music retrieving means 5 based thereon, and the music as a retrieved result is ouput by a music outputting means 6, when retrieved.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to music search, and more particularly, to a search key from music stored in a database, using a part or the whole of a music input or specified by a music input means as a search key. The present invention relates to an apparatus and a method for searching for a part or the whole of a musical piece similar to.

[0002]

2. Description of the Related Art When performing a similarity search for music, a note has been used as a matching processing unit in the conventional method. In this case, a note refers to a sequence of sounds in which the fluctuation or rate of fluctuation of the pitch falls within a certain threshold. In addition, each note has pitch information (note pitch), note length (note length), difference in pitch from a note that is temporally preceding and succeeding (relative pitch), and note information that is temporally changing as note information. It has the ratio of the note length to the note (relative note length).

FIG. 9 shows a conventional music search method.
FIG. 3 is a diagram for explaining the principle of a device using DP matching in units of musical notes. First, in the database construction phase, music to be searched is input (step S10).
1), it is converted to note information (step S102). It is stored in the database (step S103). To search for songs stored in the database, a song serving as a search key is input (step S111), and converted to musical note information (step S112). Next, the DP (Dynamic Programmable Data) for the note information of all the songs in the database and the note information obtained from the search key for the number of songs in the database.
ing) Matching is performed (step S113), and the most similar song candidates are output as search results (step S114).

As described above, when performing music similarity search, DP matching has been used in the conventional method (references 1 and 2). [Reference 1] Tomoya Sonoda, Masataka Goto, Yoichi Muraoka: "WW
Song Searching System by Singing Voice on W ", IEICE Technical Report, SP97-103, 1998. [Reference 2] Tetsuya Kageyama and Younori Takashima: “Melody Search Based on Humming Singing”, Transactions of the Institute of Electronics, Information and Communication Engineers, Vol.J77-D-II No.8 pp.1543-1551, 1994. In DP matching, a correspondence between a search key and a part or the whole of a song in the database (candidate partial tune) is determined, and a note between a search key having a correspondence and a note of a candidate partial tune is found. , The similarity is calculated by the total of the differences of the note information. DP matching searches for a candidate partial tune that maximizes the similarity with the search key by dynamic programming.

[0005]

In the conventional search method using musical notes as a unit of matching, the amount of allowable note number errors is generally limited.

In the DP matching of the conventional method, in order to suppress the amount of calculation, the difference in the number of notes between the note string of the search key and the note string of the music in the database for which the similarity is to be calculated is limited. Is provided.

[0007] Therefore, when performing music search, if a note is inserted or deleted more than a predetermined amount from a music in the database as a search target when a search key is input, search accuracy is extremely poor. There was a problem of becoming.

[0008] Even if a search key is entered,
Even if the deletion is not performed, there is a high possibility that a note is inserted or deleted when extracting the note information from the search key, and therefore, there is a problem that the search accuracy is poor.

Also, in DP matching, indexes cannot be used, and it is necessary to perform full search on all songs in the database. Even if the amount is limited, there is still a problem that search takes time.

On the other hand, in the case of a search using a music piece, the search time can be greatly reduced because the index can be used.

FIG. 10 is a diagram for explaining the principle of music search using music pieces. This method is disclosed in Japanese Patent Application No. Hei 10-34.
No. 1754 (music search device, music search method, and recording medium storing a music search program), which has been proposed by the present inventors.

First, in the construction of the database, music to be searched is input (step S201), the input music is converted into note information (step S202), and the note information is divided into a plurality of music pieces (step S201). In step S203, an index is generated so that a high-speed search can be performed (step S204), and information on the music piece is stored in the database together with the index (step S205).

Next, in the music search, a music serving as a search key is input (step S211), the input music is converted into note information (step S212), and the note information is divided into a plurality of music pieces (step S212). S213). Using each of the music pieces as a search key, a similar piece is searched from the music pieces stored in the database using the index (step S214). By examining the relationship between music pieces detected as search results and the relationship between music pieces used as search keys, the similarity between some or all of the songs in the database containing the detected music pieces is examined. The degree is checked (step S215), and part or all of the music in the database having a high degree of similarity to a part or the whole of the music used as the search key is output as a search result (step S216).

When performing a music piece search, since the number of notes of the music pieces for which the similarity is calculated matches, the DP
There was no need to find the correspondence between notes using matching, and high-speed retrieval using indexes was possible.

However, when searching for a music piece, there is a limitation that a section without insertion / deletion of a note is required for the length of the music piece because no error in the number of notes is recognized. That is, there is a problem that the search accuracy is poor unless there is a section in which notes are not inserted or deleted by the length of the music piece.

As a method that does not necessarily require note information, there is a method of performing similarity search after conversion into music information.

FIG. 11 is a diagram for explaining the principle of music search without using note information. This method is described in Japanese Patent Application
No. 329131 (music information search device, music information storage device, music information search method, music information storage method, and a recording medium recording these programs), which are proposed by the present invention.

First, in the construction of the database, music is input (step S301), and the input music is converted into music information capable of extracting a musical feature (feature amount) (step S302). It is stored in the database (step S303).

In search of music, a music serving as a search key is input (step S311), the input music is converted into equivalent music information (step S312), and the converted music information is stored in a database based on the converted music information. A similarity search is performed for the song (step S313). Here, as a similarity search, the feature amount of the music of the search key is compared with the feature amount of the music stored in the database, and the search is performed. Perform a degree calculation. Music candidates including similar ones are output as search results (step S314).

However, such a method using music information has a problem that the search accuracy is poor when the tempo between the search key and the music in the database is different.

The present invention has been made in view of the above points, and standardizes the tempo of a song and a search key in a database, so that the standardized time is used as a unit instead of a musical note. It is an object of the present invention to provide an apparatus and a method for searching for a music piece that is not affected by an error in the number of notes by enabling a search to be performed. In the present invention, since it is not necessary to use DP matching, a search using an index can be performed, and a high-speed search can be performed. Also, since the tempo is standardized before the search is performed, high search accuracy can be obtained even when the tempo between the search key and the song in the database is different.

[0022]

FIGS. 1 and 2 are diagrams for explaining the outline of a similarity search for music according to the present invention, and FIG. 3 is a block diagram showing the principle of the present invention.

According to the present invention, in music search, the tempo of music input to the database or music information converted from the music is normalized and stored in the database, and the music or the music input as a search key is stored in the database. The main feature is to perform a similarity search for music after normalizing the tempo of the converted music information.

First, in the construction of a database in a search that does not convert to music information, music is input using the music input means 1 (step S10). The input music is normalized by the tempo normalizing means 2 (step S11).

In the construction of a database in the case of a search for conversion into music information, after music is input using the music input means 1 (step S30), the music information is converted into music information by the music information conversion means 31. (Step S31),
Thereafter, the tempo is normalized by the tempo normalizing means 2 (step S32).

The tempo-normalized music or tempo-normalized music information is stored as a database in the music information storage unit 42 by the music storage means 4 (steps S12 and S33).

The tempo-normalized music or tempo-normalized music information is converted into music information by the music information conversion means 31 'and music by the music piece generation means 32 before being stored by the music storage means 4. It may be accumulated after going through the process of splitting into pieces. Furthermore, an index for accelerating the search may be created by the index generation unit 44 for the stored music, music information, and music pieces.

The music storage means 4 can store the music before the tempo standardization in the music storage section 41, and can output the music before the tempo standardization as a search result. It is possible.

In the search for the music, first, a part or the whole of the music is input or specified using the music input means 1 (steps S20, S40). The input music is normalized by the tempo normalizing means 2 (step S21), or converted into music information by the music information converting means 31 (step S41), and then normalized by the tempo (step S42). . The search key standardized in tempo may be further converted into music information by music information conversion means 31 ′ and divided into music pieces by music piece generation means 32.

The search key having undergone the above processing is sent to the music search means 5, and the music search means 5 searches for a part or all of the music similar to the search key from the music stored in the database (step S22). , S43). If an index has been generated in the database, search is performed using the index.

In the search using music pieces, the music piece search means 51 searches for similar ones from the music pieces stored in the database by using at least one of the music pieces sent as a search key. The music output means 6 outputs the music of the search result (steps S23, S44).
When searching using a plurality of music pieces sent as search keys, music pieces in a database similar to each music piece of the search key are searched, and then the music piece matching unit 52 searches for music pieces. In some cases, the consistency of the search is checked and the final search result is used.

For the matching of the music pieces, a method described in the specification of the related patent application (Japanese Patent Application No. 10-341754) is adopted. Specifically, the relationship between the music pieces detected as a search result and the relationship between the music pieces serving as the search keys of the detected music pieces are checked, and the notes before and after the detected music piece are checked. By examining the relationship between notes before and after the music piece that was the search key for the detected music piece, a part or all of the music that was the search key and the music in the database containing the detected music piece. Check the similarity between some or all of.

The music input means 1 may have a rhythm signal sending means 11 for emitting a signal at a fixed time interval or at a time interval according to a predetermined rule. In particular, for music input as a search key, tempo standardization may not be performed with sufficient accuracy. Therefore, a signal is transmitted at a fixed time interval or at a time interval according to a predetermined rule by the rhythm signal transmitting means 11, and the music inputter inputs music in accordance with the signal, and the input music and the rhythm are output. By inputting the signal transmitted by the signal transmitting means 11 to the tempo normalizing means 2 at the same time, it is possible to perform accurate tempo normalization.

Alternatively, the music input means 1 may have a rhythm signal receiving means 12. This is a means for enabling a music input user to receive a rhythm signal emitted in accordance with the performance (singing) of the music. The input music and the signal received by the rhythm signal receiving means 12 are simultaneously transmitted to the tempo normalizing means 2. , It is possible to perform accurate tempo normalization.

The tempo normalizing means 2 includes a music input by the music input means 1, music information converted by the music information converting means 31, a signal transmitted by the rhythm signal transmitting means 11, and a rhythm signal receiving means 12. Extracting means 2 for extracting the tempo from the signal received by
May have one. In this case, tempo normalizing means 2
Performs tempo normalization using the tempo extracted by the tempo extraction means 21.

The operation of the present invention will be described below. FIG.
FIG. 4 is a diagram for explaining tempo normalization. In music, the temporal length of the musical "beat" is defined by the tempo. As shown in “original music” in FIG. 4A, each music has a tempo, and this tempo may be the same or different between the music. Even the same music may be different depending on each performance. Also, the same music may be different depending on the part of the music.

Normalizing the tempo corresponds to FIG.
As shown in (B), for all performances of all songs,
In all parts, the music is converted so that the length of the time signature becomes a fixed time length.

FIG. 5 is a diagram for explaining the unit of matching. When tempo normalization is not performed, notes have been mainly used as a unit of matching. However, as shown in FIG. 5A, there is a problem that when notes are inserted or deleted, the correspondence between notes to be compared is shifted. DP matching is a method for correcting a shift in correspondence between notes to be compared when notes are inserted or deleted.

When a search is actually performed using DP matching, it is necessary to limit the amount of allowed note insertion / deletion. Therefore, note insertion / deletion greatly affects search accuracy. Further, in the search using DP matching, it is not possible to create an index for performing the search at high speed, so that a large amount of search time is required.

On the other hand, when tempo normalization is performed, FIG.
It is possible to perform matching in units of time as shown in FIG. In this case, even if a note is inserted or deleted, there is no deviation in the correspondence, and the search accuracy is not affected by that. Therefore, a highly accurate search can be performed. In addition, music whose tempo has been standardized is
Searching can be performed at high speed by dividing the music piece and creating an index.

As a technique related to the present invention, Japanese Patent Application No.
There are inventions disclosed in Japanese Patent Application No. -329131 and Japanese Patent Application No. 10-341754. The present invention, when constructing a database,
The main difference from the above-described invention is that the tempo of the music input to the database is normalized and the tempo is also normalized at the time of inputting the music serving as the search key. In particular, the present invention relates to
The main difference from the invention disclosed in Japanese Patent Application No. 0-341754 is that search is not performed in units of notes, but is performed in units of time.

[0042]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [First Embodiment] A first embodiment of the present invention will be described. MIDI data is used as music to be input to the database. At present, many songs including MIDI for karaoke are MIDI
It is circulating in the world as data.

In the MIDI data, note information is described on the basis of a time signature. The length of the beat in real time is described as tempo information separately from the note information. Therefore, if the time based on the time signature is adopted as the standardized time, the tempo of the music in the database is standardized.

When storing in the database, the original M
The IDI data is first divided into music pieces. As a method of division, the length of the music piece is made equal on the standardized time axis, and as shown in FIG.
There is a method of generating a music piece starting from the beginning of every note. Alternatively, as shown in FIG. 6B, there is a method of dividing a music piece so that the music piece starts at a fixed time interval.

After being divided into music pieces, a pitch transition vector is generated as music information (feature amount). this is,
As shown in FIG. 7, the vector is a vector obtained by engraving a standardized time axis with equally spaced scales and arranging the pitches of the sounds being played at the positions of each scale. Due to the processing by the computer, the pitch of the sound is, for example, a numerical value representing the pitch used in MIDI,
Replace with frequency, etc.

As the music information (feature amount), for example, the following can be generated in addition to the above.

The pitch transition vector A _i = (a _i1 ,
a _i2 ,..., a _in ), the average value m _i = (a _i1 + a _i2 +... + a _in ) / n of the vector elements is calculated, and the value obtained by subtracting this average value from each element is taken as the element. Vector A _i '
= (A _i1 −m _i , a _i2 −m _i ,..., A _in −m _i ) is defined as the feature value. This vector is expected to be similar even if the key of the music in the search key and the music in the database are different.

The pitch transition vector A_i= (A_i1,
a_i2, ..., a_in), The first element a _i1Based on
A vector whose elements are values obtained by subtracting this reference value from each element
A_i'' = (0, a_i2-A_i1, ..., a_in-A_i1)
And The music of the search key and the music of the music in the database
This vector is expected to be similar even if is different
Is done.

The pitch transition vector A _i = (a _i1 ,
a _i2 ,..., a _in ), a vector A _i ′ ″ = (a _i1 , a _i2 −
a _i1 ,..., a _in −a _in−1 ) are set as feature amounts. Even when the key of the music of the search key changes little by little, it is expected that this vector is similar to the music in the database.

After the music information is stored in the database, an index for speeding up the search is generated. As an index generation method, an R-tree or the like is used. The following are references related to R-tree. [Reference 3] A. Guttman, "R-trees: a dynamic index
structures for spatial searching ", Proceedings of
the ACM SIGMOD International Conference on the Man
agement of Data, pp. 47-57, Boston, 1984. There are methods of inputting a song serving as a search key, such as input by humming, input by playing an instrument, and input by playing a MIDI keyboard.

At the time of inputting music, an electronic metronome is used as a rhythm signal transmitting device. This electronic metronome can emit sound, light, characters and graphics on a computer display at a specified tempo, and the music input user can input music according to the signal emitted. It has become. This electronic metronome not only emits signals at regular time intervals, but also, for example, the tone at the beginning of a bar
Rhythm according to the type of music to be input, such as the color of light, the intensity of light, the type of characters and the type of graphics on the display, which are different from other beats, and the type of song to be input, such as "Samba rhythm" A signal can be emitted. The tempo at which this signal is emitted can be freely set to be slower or faster so that the music inputter can easily input the music.

Alternatively, when inputting music, a rhythm signal receiving device may be used. This is clapping, stepping,
The music enthusiast is conscious / unconscious when inputting music, such as finger up / down movements, body shaking, mouse clicks, keyboard presses, etc.
It is a device that can detect unintentional movements, signals, etc.

Using the input music and the signal obtained from the rhythm signal receiving / transmitting device, the tempo of the music serving as a search key is normalized.

The division into music pieces and the conversion to music information are the same as those for music input to the database for music input from the MIDI keyboard.

In the input by humming and the input by playing a musical instrument, the music input means 1 inputs a music as PCM format data. Then, pitch information is extracted from this data. The pitch information is extracted according to the conventional method of extracting note information from PCM format data. [Reference 4] Takahara Niihara, Masakazu Imai, Seiji Iguchi: "Automatic Transcription of Singing", Transactions of the Society of Instrument and Control Engineers, Vol.20 No.10
pp.940-945, 1984; [Reference 5] Masanori Mizuno, Masaki Fujimoto, Yonori Takashima, Shichiro Tsuruta: "Personal Computer Music System-Automatic Transcription of Singing Voices", Information Processing Society of Japan 35th Annual Convention, 5Ff- 5, 1987. That is, first, frequency analysis is performed using FFT or autocorrelation processing, and peak extraction, power measurement of each peak, and the like are performed. Further, noise is removed, pitch drift is corrected, harmonic components are absorbed into basic components, and the like, and which sound is being played at which point on the time axis is detected. The method of dividing into musical pieces and generating the music information after the pitch information is extracted is the same as the case of inputting music to the database.

It should be noted that the music information generation process may be divided into two stages, and the tempo may be normalized during the process. For example, after extracting pitch information, normalizing the tempo,
In some cases, a feature amount such as a pitch vector is generated.

As for the search, first, a music piece similar to each of the music pieces generated from the music input as the search key is searched from the database (music piece search). At this time, the distance between the music piece of the search key and the music piece in the database is obtained for each feature amount such as a pitch vector. This distance is obtained by setting an appropriate distance such as the Euclidean distance or the Manhattan distance for each feature amount. Also, in order to speed up the search, the search is performed using the index.

The distance (dissimilarity) between music pieces is expressed as a value obtained by some method by adding an appropriate weight to the distance of each feature value. The details are as follows. A set of used feature amounts is set to FSPACE. The distance of the feature piece i between the music piece K and the music piece X is represented by di (K, X),
Assuming that the weight is Wi, obtaining the total distance D (K, X) between the music piece K and the music piece X by some method is performed using the following function F.

D (X, K) = F (｛Wi｝, ｛di (K, X)｝) (where i∈FSSPACE) As a specific example of this function F, if a function of linear sum is used , D (X, K) = sum (Wi × di (K, X)) (where i∈FSSPACE, and sum is i
を Sum the total for FSPACE).

As the function F, instead of a linear sum, for example, a function that takes the following minimum value can be used.

D (X, K) = min (Wi × di (K, X)) (where i∈FSPAC, min is i∈FSPAC)
(The minimum value for E is taken.) A mixture of the above linear sum and the minimum value can also be used. That is, two types of feature values (FSSPACE) (FS
PSPACE1, FSPACE2), and if FSPACE = FSSPACE1∪FSSPACE2, then D (X, K) = sum (Wi × di (K, X), min
The total distance can also be obtained using a function such as (Wj × dj (K, X)).
Here, i∈FSSPACE1, j∈FSSPACE2.

Here, the distance between music pieces (dissimilarity)
As an example of the method for obtaining the total distance, a case where a linear sum is used, a case where a minimum value is used, and a case where a mixture of the linear sum and the minimum value are used have been described. Of the music pieces in the database, a music piece whose distance to the music piece of the search key is short, that is, a music piece with a high similarity is a result of the music piece search.

When a plurality of music pieces are generated from the search key, the consistency between the music pieces is checked based on the result of the music piece search for each of the music pieces of the search key, and the similarity when a plurality of music pieces are combined is determined. The degree can be calculated and used as the final search result.

Now, music pieces K _p , K _q (p <
to q), the music piece X in the database, Y was similar, respectively _{(K p ~X, K q ~Y} ). Music piece X, Y are, that have the consistency the music piece K _p, K _q of the search key is that the following conditions are satisfied.

1) Music pieces X and Y must belong to the same music. Incidentally, the candidate music and A is later put music piece X, Y, respectively A _r, and A _s.

2) Music piece A_r, A_sThe temporal order of
The music fragment (K_p, K _q) And the temporal order
The same (ie, r <s).

[0067] 3) Music piece A _r, the starting time in the time that is normalized A _s, T _Ar, and T _As, music piece K _p of the search key, a start time in the normalized time K _q ,
When T _Kp and T _Kq are set, the intervals of the start times match, that is, T _As −T _Ar = T _Kq −T _Kp .

In the output of the search result, the similarity between the search key and a part or the whole of the music in the database which is a candidate for the search result is obtained.
Present part or all of the song. The title of the music and the name of the musician can be displayed from the information in the database, and a part or the whole of the music similar to the search key can be played.

In the present embodiment, MIDI data can be used as music input to the database. MIDI data is in a standardized format, and a large number of music pieces exist as MIDI data, so that it is not necessary to perform music pieces for database construction, and database construction is easy. Since MIDI information describes note information on the basis of a time signature, tempo standardization and pitch information extraction can be accurately performed, so that the MIDI data is suitable for use as a database for music search.

Also, when inputting a music piece serving as a search key,
By using the rhythm signal receiving / sending device as an auxiliary device, the tempo of a music piece serving as a search key can be accurately standardized.

The special effects of the present embodiment include that the database can be easily and accurately constructed, and that the tempo of the music used as the search key can be accurately standardized.

[Second Embodiment] In the second embodiment of the present invention, MIDI data is not used as music input to the database in the first embodiment. (Compact disc) data, songs that have been PCM-encoded in a format such as WAVE, AIFF, or AU, or if they are in another encoding format, have been once converted to PCM-encoding format Use something.

Generally, in music, a sound is played in time with the beat. Since the sound greatly changes in sound pressure at the moment when the performance is started, it is possible to extract tempo information by examining the change in sound pressure in the music.

As a more specific method of extracting tempo information,
There are the following methods.

A method of searching for a peak of a volume level at a small sampling frequency such as 100 Hz or 10 Hz, and extracting a tempo from the interval between the peaks. Assuming that the sampling frequency of the original PCM data is, for example, 44100 Hz, a method of creating data in which 1000 samples are added together and searching for the peak of the volume level can be considered.

A method of examining the transition of the rate of change of the volume level at the small sampling frequency, searching for a portion where the change rate sharply increases, and extracting the tempo from the interval of the rapidly increasing portion.

A method of performing frequency analysis on the data of the small sampling frequency in a long section, for example, 10 seconds, and searching for a peak. For example, if a peak appears at 2 Hz, it means that the length of one beat of this music piece is 0.5 seconds.

A method for extracting tempo information by selectively extracting sounds of musical instruments such as drums that carve a rhythm and analyzing fluctuations in sound pressure of these sounds.

As a method of generating a feature amount such as pitch information from an input music, the method of extracting pitch information or the like from a music input by humming or playing a musical instrument in the first embodiment is described. According to. However, unlike humming and the performance of a single instrument, data such as music CDs have multiple notes emitted simultaneously, so the pitch to be searched (the pitch of the sound corresponding to the melody) Only need to extract. For this purpose, after FFT or autocorrelation processing, a method of extracting the peak of the sound pressure, or a bandpass filter that extracts only the peak of the frequency band of a musical instrument playing melody or rhythm or a human singing voice There is a way to use

The effects of the present embodiment are as follows. Enhance the amount of songs contained in the database by using songs encoded by the PCM method, which is a more general music encoding method than MIDI, as songs input to the database. Is possible. Further, the database can include not only commercially available music CDs, but also songs played personally by musical instruments.

[Third Embodiment] In the third embodiment of the present invention, the music in the database and the music of the search key in the first and second embodiments are divided into music pieces. When generating the feature amount, instead of generating a pitch transition vector, a distribution of a pitch in a target music piece is obtained, and a feature amount vector is generated based on this distribution.

In a certain music piece, the pitch is as shown in FIG.
When a transition is made as shown in Fig. 8, one unit of the standardized time is converted into a number once, and a pitch distribution diagram as shown in Fig. 8B is created, and each pitch (or range of pitch) is created. The pitch distribution vector can be created by arranging the frequencies of. This can be searched as music information (feature amount).

As the music information (feature amount), for example, the following can be generated in addition to the above.

The distribution of element values of the vector A _i ′ obtained by subtracting the average value of the elements from each element of the pitch transition vector A _i used in the first embodiment.

The distribution of element values of the vector A _i ″ obtained by subtracting the value of the first element from each element of the pitch transition vector A _i used in the first embodiment.

The distribution of element values of the vector A _i ′ ″ obtained by subtracting the value of the previous element from each element of the pitch transition vector A _i used in the first embodiment.

In each of the above distributions, the frequency of the portion where the value corresponding to the horizontal axis of the pitch distribution diagram of FIG. 8B is 0 may be set to 0. In each of the above distributions,
When the value corresponding to the horizontal axis of the pitch distribution diagram of FIG. 8B is limited to a predetermined range and data outside the range is deleted, or added to the value at the boundary of the range. is there.

Using these feature amount vectors, similar search is performed after music piece search and music piece matching according to the method of the first embodiment.

Regarding the effect of this embodiment, since the number of dimensions of the feature vector can be generally reduced as compared with the case of the first embodiment, the search speed can be improved. In addition, since the time information is lost within the range of the music piece, even if there is some error between the music piece in the database and the start point of the music piece generated from the search key, the search can be performed. . On the other hand, since the time information is lost, even if they are not similar in the pitch transition vector, they may be similar in the pitch distribution vector. Therefore, it is also possible to first perform a search by the method according to the present embodiment, and then perform a narrowed-down search for performing a search according to the first embodiment from the search results.

[0090]

As described above, according to the present invention, the searcher can freely input a part or the whole of the music by humming or playing a musical instrument, and the music intended by the searcher can be stored in the database. It is possible to search at high speed from within.

According to the present invention, for example, in a system such as a karaoke system or a music-on-demand system (automatic jukebox), when the user cannot remember the name of the music to be selected, the name of the composer, the name of the performer, etc. However, if the melody of the music is memorized, the target music can be selected, which is extremely effective in such a field.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating an outline of a similarity search for music (when conversion to music information is not performed) according to the present invention.

FIG. 2 is a diagram illustrating an outline of a similarity search for music (when conversion to music information is performed) according to the present invention.

FIG. 3 is a diagram illustrating the principle of the present invention;

FIG. 4 is a diagram illustrating tempo normalization.

FIG. 5 is a diagram illustrating a unit of matching.

FIG. 6 is a diagram illustrating division into music pieces.

FIG. 7 is a diagram illustrating conversion into music information.

FIG. 8 is a diagram illustrating conversion into music information (pitch distribution).

FIG. 9 is a diagram illustrating the principle of music search (using DP matching) according to a conventional method.

FIG. 10 is a diagram illustrating the principle of music search using a music piece related to the present invention.

FIG. 11 is a diagram illustrating the principle of music search that does not use note information according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Music input means 11 Rhythm signal transmission means 12 Rhythm signal reception means 2 Tempo normalization means 21 Tempo extraction means 31 Music information conversion means 31 'Music information conversion means 32 Music piece generation means 4 Music storage means 41 Music storage part 42 Music information Storage unit 43 Music piece storage unit 44 Index generation means 5 Music piece search means 51 Music piece search means 52 Music piece matching means

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Tetsuo Sakata 2-3-1 Otemachi, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (72) Inventor Masashi Yamamuro 2-3-1, Otemachi, Chiyoda-ku, Tokyo No. 1 Within Nippon Telegraph and Telephone Corporation (72) Masayoshi Umeda 2-3-1 Otemachi, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (72) Inventor Seiichi Konya Otemachi, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation

Claims (14)

[Claims] In a search device for searching for music, a music input means for inputting music, a tempo normalizing means for normalizing the tempo of the music input by the music input means, and at least the tempo normalizing means. A music storing means for storing music whose tempo has been standardized, and a part or the whole of the music inputted as a search key, based on a part or the whole of the music tempo standardized by the tempo standardizing means, A music search means for searching for a similar music from a part or the whole of the music stored in the music storage means; and a music output means for outputting the music searched by the music search means. Music search device. 2. The music search device according to claim 1, wherein
Music information conversion means for converting the music input by the music input means into music information, wherein the tempo normalization means performs tempo normalization after converting the music into music information by the music information conversion means; A music search device characterized by the above-mentioned. 3. The music search device according to claim 1, wherein the music input means transmits a signal at a fixed time interval or at a time interval according to a predetermined rule, or A music search device comprising a rhythm signal receiving means for allowing a music input user to input a rhythm of a music being input. 4. The music search device according to claim 1, wherein said tempo normalizing means comprises:
Using at least one of the music input by the music input means, the music information converted by the music information conversion means, the signal transmitted by the rhythm signal transmission means, and the signal received by the rhythm signal reception means. A music search device comprising tempo extraction means for extracting a tempo, and performing tempo normalization using the extracted tempo. 5. The music search device according to claim 1, wherein the music search means is based on a part or the whole of a search key standardized in tempo. When searching for part or all of the tempo-normalized music stored in the tempo, the same as part or all of the tempo-normalized search key and part or all of the tempo-normalized search key A music search device characterized by examining the degree of similarity with a part or the whole of a tempo standardized music to be searched. 6. The music retrieval device according to claim 1, wherein the music whose tempo is normalized or the music information whose tempo is normalized is divided into music pieces. A music search device comprising a piece generating means. 7. The music search device according to claim 6, wherein
The music piece generating device is characterized in that the music piece generation means makes the length of all music pieces constant when generating music pieces. 8. The music search device according to claim 6, wherein the music piece generating means starts the music piece for the same music piece at a constant time interval when generating the music piece. Music search, wherein the music piece is generated from the beginning of a note, or from the point where the fluctuation or rate of the pitch exceeds a certain threshold. apparatus. 9. The music search device according to claim 1, further comprising music information conversion means for converting the music information into music information before storing the music information in the music storage means. Music search device. 10. The music search apparatus according to claim 1, further comprising: an index generation unit configured to generate an index for speeding up a search when storing music. A music search device, wherein the search means uses an index when performing a search. 11. The music search device according to claim 6, wherein said music search means comprises:
A music search device comprising: a music piece search unit that searches for music pieces when performing a search and derives a final search result from a relationship between the resulting music pieces. 12. The music search device according to claim 11, wherein the music search means checks the consistency between the relationship between the music pieces of the search key and the relationship between the music pieces of the search result.
A music search device comprising music piece matching means for leading a final search result. 13. A search method for searching for a song, a process of inputting a song, a process of normalizing the tempo of the input song, a process of storing at least the tempo-normalized song, and Searching for similar music from a part or the whole of the stored music based on a part or the whole of the tempo-normalized music obtained from a part or the whole of the input music; And outputting the searched music. 14. A program recording medium in which a program used for searching for a song by a computer is recorded, wherein the process of inputting the song, the process of normalizing the tempo of the input song, and the process of inputting as a search key are performed. Based on a part or the whole of the tempo-normalized music obtained from a part or the whole of the music that has been set, the tempo-normalized part of the whole or part of the music to be searched is stored. A recording medium on which a music search program is recorded, wherein a program for causing a computer to execute a process of searching for a similar song and a process of outputting the searched song is recorded.