JP4241445B2 - Music content search device, program, and computer-readable recording medium - Google Patents

Description

The present invention relates to a musical content search device及beauty program and a computer-readable recording medium, in particular, music contents for searching the music content using the characteristic portion in the input data retrieval apparatus及beauty program and a computer-readable The present invention relates to various recording media .

In conventional music search, music is divided into pieces of music in advance and assigned with importance, and a part or all of the music is used as a search key to search for similar music based on importance. There are methods (see, for example, Patent Documents 1 to 4 and Non-Patent Documents 1 and 2).
Japanese Patent Laid-Open No. 2001-109471 “Music Retrieval Device, Music Retrieval Method, and Storage Medium Recording Music Retrieval Program” Nishihara, Kosugi, Sugaya, Yamamuro JP 2002-175073 A "Performance Collection Device, Performance Collection Method and Performance Collection Program Recording Medium" Kataoka, Kosugi, Kojima JP 2000-347659 A “Music Search Device, Music Search Method, and Recording Medium Recording a Music Search Program” Nishihara, Kosugi, Sakata, Yamamuro, Umeda, Kajitani JP 2002-123287 A “Music Feature Generation Method and Device, Music Information Retrieval Device, and Music Feature Generation Program Recording Medium” Naoko Kosugi "Humming search system for large-scale music database" IPSJ Journal 2002 February "Sound Compass: Music Search System by Humming" IPSJ Journal, January 2004

  However, in music content search, it is important to see how effectively input data can be used. Partial analysis of music has also revealed that there are similar melodies that are used in multiple songs. Such a portion has a poor feature from the viewpoint of search and causes a reduction in search resolution.

The present invention has been made in view of the above, and an object thereof is to provide a musical content search device及beauty program and a computer-readable recording medium capable of realizing high resolution.

  FIG. 1 is a principle configuration diagram of the present invention.

The present invention is a music content search device for searching for songs stored in the music database 30,
When music data is input, a part of the music data is divided into a plurality of partial music data, and create multiple subqueries from partial music data, a single query are collectively subquery plurality of one Subquery creation means 10 to create;
The music database 30 is searched by a query, and for each song stored in the music database 30, the minimum distance between the song and each partial music data is defined as the distance between the song and the input music data. Search means 20 for finding the top n songs with the smallest distance ;
If the ratio of the partial music data that gave the minimum distance for each of n pieces of music to the search result searched for by the search means 20 is greater than or equal to a predetermined value, it becomes equal to or higher than the predetermined value. An effective subquery detection means 40 for excluding subqueries corresponding to partial music data and combining the remaining subqueries into one to create one query;
A final result shaping / processing unit 50 which shapes the result of re-searching the music database 30 by the search unit 20 using the query generated by the valid subquery detection unit 40 and outputs the result as a final search result. .

  Further, the subquery creation means 10 of the present invention includes means for generating partial music data of a certain length while shifting at a certain interval when dividing a part of music data into a plurality of partial music data. .

  FIG. 2 is a diagram for explaining the principle of the present invention.

The present invention is a music content search method for searching for songs stored in a music database,
When music data is input (step 1), a part of the music data is divided into a plurality of partial music data (step 2), to create multiple subqueries from partial music data, 1 subqueries plurality of Creating a single query in one step (step 3) subquery creation step;
The music database is searched by query, and for each song stored in the music database 30, the distance between the song and each piece of music data is defined as the distance between the song and the input music data. and search steps (step 4) to seek the song of a small top n,
If the proportion of the partial music data that gives the minimum distance for each of n songs to the search result searched in the search step is greater than or equal to a predetermined value, the portion that is equal to or greater than the predetermined value exclude subquery corresponding to the music data, the effective subquery detection steps to create a single query are collectively remaining subquery to one (step 5),
Effective by the search means using a subquery queries generated by the detection step performs search of music database (step 6), and shaping processing result is output as a final search result final results shaping and processing steps (step 7) .

The present invention is a program for causing a computer to function as each means constituting the music content search device according to claim 1 or 2.

The present invention is a computer-readable recording medium storing the program according to claim 3.

  As described above, according to the present invention, a portion with poor features is detected from input data and the portion is not used for search. Specifically, a plurality of subqueries based on the portion from inquiry music data are used. And then search the music database with a query consisting of those subqueries, analyze the search results from those subqueries, detect subqueries that are considered valid for the search, and use the query consisting only of them again The music database is searched and the result is shaped and processed as the final result.

  Specifically, for example, it is characterized in that a search is performed centering on a more characteristic part in the input Hamming data, and one tune is composed of a very characteristic melody. When a part (such as “rust”) and a part composed of a melody that is often used in relatively many songs (such as the end of a song) are mixed, a number of songs are included in the humming search. If there is a melody like that used in the search, a search using that part will output a search result saying that many songs are correct, limiting which song is actually hummed. You can't. In the present invention, the search is performed by eliminating a portion having such a poor feature, thereby limiting the hummed music more accurately.

  According to the present invention, it is possible to automatically determine a pitch and a sound value from an acoustic signal and tempo information of an input person's singing voice (singing with a tatter or lyric). Further, by performing preprocessing before frequency analysis and determining the necessity / unnecessity of time-consuming frequency analysis, the processing speed can be increased.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 3 shows a configuration of a music content search apparatus according to an embodiment of the present invention.

  The music content search apparatus shown in FIG. 1 includes a subquery creation unit 10, a search unit 20, a music database (DB) 30, an effective subquery detection unit 40, and a final result shaping unit 50.

  When a part of music data is input as input information, the sub-query creating unit 10 divides a part of the music data into a plurality of sub-data, creates a sub-query from each divided sub-data, and collects the sub-queries. One query is created and passed to the search unit 20.

  The search unit 20 searches the music DB 30 with the subquery created by the subquery creation unit 10 and sends the search result to the valid subquery detection unit 40. In addition, the music DB 30 is searched again using the query acquired from the valid subquery detection unit 40, and the search result is sent to the final result shaping unit 50.

  The valid subquery detection unit 40 analyzes the search result acquired from the search unit 20, detects a valid subquery, creates a query using only the valid subquery, and returns the query to the search unit 20.

  The final result shaping unit 50 shapes and processes the search result re-searched by the query in the search unit 20 and outputs it as the final result.

  Next, the operation in the above configuration will be described.

  FIG. 4 is a flowchart of a series of operations according to the embodiment of the present invention.

  Step 101) When music data is input, the sub-query creating unit 10 divides the music data into parts to generate partial music data. At this time, as shown in FIG. 5, a plurality of pieces of partial music data having a length (window length) being cut are extracted while being shifted at a constant interval (slide width). In this case, the sliding window method (C. faloutsos, M. Ranganathan, and Y. Manolopoulos. Fast Subsequence Matching in Time-Series Database. In Proceedings of the ACM SIGMOD, International Conference on management of Data, pp.419 -429, 1994).

Step 102) The subquery creation unit 10 creates a subquery from the partial music data. At this time, feature data (for example, feature vector) is extracted and used as a subquery. For example, as the feature data, when music data is divided at regular time intervals, the pitch of the sound that is reproduced for the longest time is extracted. For example, as a method for extracting feature vectors, `` N. Kosugi, Y. Nishihara, T. Sakata, M. Yamamuro, and K. Kushima, A Practical Query-By-Humming System for a Large Music Database.In Proc, of ^{the 8 th ACM International Conference on Multimedia} , pp. 333-342, it is possible to use a 2000 ".

  Step 103) The subquery generator 10 combines the subqueries into one query (for example, an SQL statement).

  Step 104) The search unit 20 searches the music DB 30 using the query created in step 103. The search is a similar search (for example, Naoko Kosugi, Yoshi Sakurai, Masashi Yamamuro, Kazuhiko Kushima, Sound Compass: “Music Search System by Hamming”, Information Processing Society Journal, Vol. 45, No. 1, Jan, 2004). Similarity search is, for example, when a numeric string is entered, the distance between the numeric string and each numeric string in the database is calculated, and the top few items with the closest distance are used as search results. is there. As a search result, the song title and its similarity are acquired and transferred to the valid subquery detection unit 40.

  Step 105) The valid subquery detection unit 40 determines whether or not there is a search result including the acquired song name and its similarity that matches a preset condition. If there is, the process proceeds to step 106, and if not, the process proceeds to step 107.

  Step 106) If there is a subquery that satisfies the condition, the valid subquery detection unit 40 excludes the subquery and collects the remaining one subquery (for example, an SQL statement). For example, if four subqueries are created from one hamming and the fourth subquery meets the condition, a query is created using only the subqueries 1 to 3, passed to the search unit 20, and the process proceeds to step 104.

  Step 107) If there is no subquery that matches the condition in Step 105, the final result shaping unit 50 shapes and processes the search result. For example, as shown in FIG. 6, in the case of a WWW browser, it is processed into an HTML format. As a result of shaping and processing, for example, a list of song names arranged in order of similarity and the similarity of each station are output as the final search result.

  Next, the operation of the valid subquery detection unit 40 in step 105 will be described in detail.

For example, when music data h is input to a music DB 30 that holds a song a _k (1 ≦ k), and a music content search is performed to output a song list of the top n songs with high similarity as a final search result. To do.

Let m be the number of partial music data created from the input music data h, and p be the number of partial music data for the song _ak . The distance between the i-th partial music data of the song _{ak and} the j-th partial music data of the input music data

としたとき、曲ａ_ｋと入力音楽データｈとの距離Ｄ（ａ_ｋ，ｈ）を、

If the distance D (a _k , h) between the music a _k and the input music data h is

とする。このとき、Ｄ（ａ_ｋ，ｈ）の値が小さい上位ｎ件が最終結果となるが、ｎ個のＤ（ａ_ｋ，ｈ）を算出したｈ_ｊのそれぞれの比率を、そのｈ_ｊの最終結果決定率

And At this time, the top n cases where the value of D (a _k , h) is small is the final result, and the ratio of each of h _j from which n D (a _k , h) is calculated is the final value of h _j Result decision rate

と定義し以下の式で算出する。

And calculate with the following formula.

ここで、

here,

がある閾値を越えて大きかった場合（例えばｘ％を越えた場合）、ｈ_ｊは入力音楽データの中で特徴が乏しい部分であると判断する。

If the threshold value is larger than a certain threshold value (for example, if it exceeds x%), h _j is determined to be a portion with poor features in the input music data.

Therefore, for example, as a condition,
“Final result decision rate exceeds x%”
It can be. For example, x may be “80”.

  The above operation is shown in FIG.

  As described above, once the music database 30 is searched with a query, the search result is analyzed, and if a portion with poor features is included in it, a query that excludes that portion is generated and searched again. By using the result as the final search result, high resolution can be realized.

  4 and 7 in the above embodiment is constructed as a program, installed in a computer used as a music content search device, executed by a control means such as a CPU, or distributed through a network. It can also be installed on a computer.

  Further, the constructed program can be stored in a hard disk connected to a computer used as a music content search device or a portable storage medium such as a CD-ROM and installed in the computer.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

  The present invention can be applied to search for music content and the like and music.

It is a principle block diagram of this invention. It is a figure for demonstrating the principle of this invention. It is a block diagram of the music content search apparatus in one embodiment of this invention. It is a flowchart of a series of operation | movement in one embodiment of this invention. It is a figure which divides | segments the input music data in one embodiment of this invention based on a part. It is an example of shaping and processing of a final search result in an embodiment of the present invention. It is a flowchart of the search result analysis process in one embodiment of the present invention.

Explanation of symbols

10 subquery creation means, subquery creation section 20 search means, search section 30 music DB (database)
40 Effective subquery detection means, effective subquery detection section 50 Final result shaping / processing means, final result shaping / processing section

Claims (4)

A music content search device for searching for songs stored in a music database,
When music data is input, a part of the music data is divided into a plurality of partial music data, and create multiple subqueries from partial music data, a single query are collectively subquery plurality of one A subquery creation means to create,
The music database is searched by the query, and for each song stored in the music database, the minimum distance among the songs and the partial music data is set as the distance between the song and the input music data. A search means for finding the top n songs with a short distance ,
If the ratio of the partial music data that gives the minimum distance for each of n pieces of music to the search result searched for by the search means is greater than or equal to the predetermined value, it becomes equal to or higher than the predetermined value. Effective subquery detection means for excluding subqueries corresponding to the partial music data and creating one query by combining the remaining subqueries into one ,
A final result shaping / processing means for shaping the result of re-searching the music database by the search means using the query generated by the effective subquery detection means, and outputting the result as a final search result;
A music content search apparatus comprising: The subquery creation means includes:
2. The music content search apparatus according to claim 1, further comprising means for generating partial music data having a predetermined length while shifting at a predetermined interval when dividing a part of the music data into a plurality of partial music data. The program for functioning a computer as each means which comprises the music content search apparatus of Claim 1 or 2 . A computer-readable recording medium storing the program according to claim 3 .

Images