CN101673262B - Method for searching audio content - Google Patents

Abstract

The invention discloses a method for searching an audio content, which comprises:1) an audio fingerprint extraction step, which is to extract audio fingerprints of a plurality of audio files; 2) an audio fingerprint segmentation step, which is to segment the audio fingerprints extracted by the step 1); 3) an index generation step, which is to generate an audio fingerprint index according to the result of the segmentation in the step 2); and 4) a searching step, which is to search a matched audio file by using the audio fingerprint index. In the method, a segmentation technique in a text search engine is used to perform segmentation processing of the audio fingerprint files first, then an index technique in the field of text search is used to perform the index processing of the audio fingerprints, and finally, after the index processing is completed, the search engine can search an audio segment input by a user. Thus, the method facilitates the search of the user and improves search efficiency.

Description

The searching method of audio content

Technical field

The present invention relates to the searching method of audio content.

Background technology

Along with Internet development, search engine becomes people's one of necessary tool of surfing the Net.Traditional search engine all is based on text search (Text Search), and being called is text search engine.Its principle is: search engine server is collected a large amount of webpages; And according to the text in the existing Rule Extraction webpage and do participle (Word Segmentation) and handle, common segmenting method, for example: based on the segmenting method of string matching, based on the segmenting method of understanding with based on the segmenting method of statistics; Text search engine utilizes the text dictionary to index and shows to be used for quick search.The user is input to server with text when searching for, server is searched for according to concordance list, then return results after the text is carried out word segmentation processing fast.

At present, search engine all is based on text, searches for even the search engine of some search pictures or audio frequency also is text messages such as title, explanation, introduction, label through picture or audio program.Search engine does not also have directly to search for through the signal content of audio frequency.

Audio-frequency fingerprint (audio fingerprinting) just is being suggested a long time ago; For example, Jaap Haitsma and TonKalke have delivered " a kind of audio fingerprint system of high reliability " (A Highly Robust AudioFingerprinting System) on music searching in 2002 makes progress international conference (Proceedings of International Conference on MusicInformation Retrieval).This system passes through method for processing signals; With the sound signal of (for example 11.6ms) at set intervals in the audio file; Be converted into the fingerprint (fingerprint) of one 32 bit (bit) size, an audio file just can be converted into a file fingerprint by this method.System just can carry out fast audio-frequency fingerprint and retrieve behind table that all audio-frequency fingerprint files are indexed.

Under audio-frequency fingerprint number of files less (for example 10,000 s') situation, can all file fingerprints be deposited in the calculator memory, carry out index after, can retrieve fast easily.Above-mentioned " a kind of audio fingerprint system of high reliability " promptly provided the detailed step of this method.Yet under actual conditions, the number of audio file will be considerably beyond 10,000 number.For example, the number of audio files that occurs on the internet at present surpasses 10,000,000 numbers, and quantity is in continuous growth.Therefore adopt this method to be difficult to make practical search engine.

Summary of the invention

In order to solve the problems of the technologies described above, the present invention provides a kind of searching method of audio content, and it is audio-frequency fingerprint search engine (audio fingerprint search engine) that this search engine is called.

The present invention adopts following technical scheme:

A kind of searching method of audio content comprises:

1) audio-frequency fingerprint extraction step extracts the audio-frequency fingerprint of a plurality of audio files;

2) audio-frequency fingerprint participle step, the audio-frequency fingerprint that step 1) is extracted carries out participle;

3) index generates step, according to step 2) word segmentation result generate the audio-frequency fingerprint index;

4) search step, the audio file that utilizes this audio-frequency fingerprint indexed search to mate.

Wherein, said step 4) specifically may further comprise the steps:

According to audio file or the audio file fragment that the needs of input are retrieved, extract its audio-frequency fingerprint, this audio-frequency fingerprint is carried out participle, according to the audio file of word segmentation result search matched in said audio-frequency fingerprint index.

Wherein, further comprising the steps of before the said step 4) after the said step 3):

Storing step, store audio fingerprints, said audio-frequency fingerprint index and corresponding audio files thereof.

The present invention is through adopting the participle technique in the text search engine; On the audio-frequency fingerprint file, carry out word segmentation processing; Adopt the index technology in the text search field that audio-frequency fingerprint is carried out index process then; After index process was accomplished, search engine can be searched for the audio fragment of user's input.Not only make things convenient for user's search, and improved the efficient of search.

Further specify the present invention below in conjunction with accompanying drawing and embodiment.

Description of drawings

Fig. 1 is the searching method embodiment schematic flow sheet of audio content of the present invention.

Embodiment

As shown in Figure 1, a kind of searching method of audio content comprises:

1) audio-frequency fingerprint extraction step extracts the audio-frequency fingerprint of a plurality of audio files;

2) audio-frequency fingerprint participle step, the audio-frequency fingerprint that step 1) is extracted carries out participle;

3) index generates step, according to step 2) word segmentation result generate the audio-frequency fingerprint index;

4) search step, the audio file that utilizes this audio-frequency fingerprint indexed search to mate.

Wherein, said step 4) specifically may further comprise the steps:

According to audio file or the audio file fragment that the needs of input are retrieved, extract its audio-frequency fingerprint, this audio-frequency fingerprint is carried out participle, according to the audio file of word segmentation result search matched in said audio-frequency fingerprint index.

Wherein, further comprising the steps of before the said step 4) after the said step 3):

Storing step, store audio fingerprints, said audio-frequency fingerprint index and corresponding audio files thereof.

Participle mode in the foregoing description can adopt multiple mode to realize, below enumerates several kinds of modes and explains respectively.

Mode one

Employing is carried out word segmentation processing based on the Statistic for Chinese segmenting method to audio-frequency fingerprint.At first with the file fingerprint of 15000 audio files by the method generation fixed width of above-mentioned Jaap Haitsma and Ton Kalke, its width can be 32 bits or 16 bits, and each file fingerprint that obtains on average is made up of the fingerprint of about 10000 fixed width.The data of each 32 bit or 16 bits are counted as a word in the Chinese.It is 15000 pieces " articles " that all 15000 file fingerprints that comprise " word " are taken as, and these " articles " then carry out participle as the language material of Chinese word segmentation.In statistic processes, the frequency of the combination of each " word " of adjacent co-occurrence in the audio frequency language material is added up.The combination that the co-occurrence frequency is high is considered to a speech, is called " fingerprint speech ".For example; The combination of the fingerprint of 7 continuous scale-of-two " 00000000000000000000000000000000 " that frequency is higher; With the combination of the fingerprint of 5 continuous scale-of-two " 11111111111111111111111111111111 " be the higher fingerprint combination of frequency by statistics, they are used as " fingerprint speech ".

Mode two

Adopting the fingerprint width is the audio-frequency fingerprint method for distilling of 16 bits.Specifically be to be that the fingerprint of 32 bits carries out the fingerprint that interval sampling obtains 16 bits with width in the mode one.Adopt identical with mode one word segmentation processing of carrying out audio-frequency fingerprint based on the Statistic for Chinese segmenting method then.

Claims (3)

1. the searching method of an audio content is characterized in that comprising:

1) audio-frequency fingerprint extraction step extracts the audio-frequency fingerprint of a plurality of audio files;

2) audio-frequency fingerprint participle step, the audio-frequency fingerprint that step 1) is extracted carries out participle;

3) index generates step, according to step 2) word segmentation result generate the audio-frequency fingerprint index;

4) search step, the audio file that utilizes this audio-frequency fingerprint indexed search to mate;

Said audio-frequency fingerprint participle may further comprise the steps:

A. audio file generates the file fingerprint of fixed width, and its width is 32 bits or 16 bits;

B. the frequency of each file fingerprint of adjacent co-occurrence in the audio file is added up;

C. the combination that the co-occurrence frequency is high is considered to a speech and carries out.

2. the searching method of audio content according to claim 1 is characterized in that, said step 4) specifically may further comprise the steps:

According to audio file or the audio file fragment that the needs of input are retrieved, extract its audio-frequency fingerprint, this audio-frequency fingerprint is carried out participle, according to the audio file of word segmentation result search matched in said audio-frequency fingerprint index.

3. the searching method of audio content according to claim 1 and 2 is characterized in that, and is further comprising the steps of before the said step 4) after the said step 3):

Storing step, store audio fingerprints, said audio-frequency fingerprint index and corresponding audio files thereof.

Images