WO2008127052A1

WO2008127052A1 - System and method for searching audio fingerprint by index information

Info

Publication number: WO2008127052A1
Application number: PCT/KR2008/002085
Authority: WO
Inventors: Seungjae Lee; Jin Soo Seo; Sang Kwang Lee; Wonyoung Yoo; Young Suk Yoon; Yong Seok Seo; Weon Geun Oh; Young Ho Suh
Original assignee: Electronics And Telecommunications Research Institute
Priority date: 2007-04-17
Filing date: 2008-04-14
Publication date: 2008-10-23
Also published as: CN101663708A; KR100862616B1; CN101663708B

Abstract

Provided are a system and method for searching audio fingerprint by index information. The system includes a DB group for generating an index based on statistical characteristics of an audio fingerprint for an audio file and consecutively matching the index, the audio fingerprint, and music information, and an audio fingerprint searching apparatus for generating a new index based on statistical characteristic of an audio fingerprint for a new input audio file and searching corresponding music information for the new input audio file by searching the new index from the DB group.

Description

SYSTEM AND METHOD FOR SEARCHING AUDIO FINGERPRINT BY INDEX INFORMATION

Technical Field

[1] The present invention relates to an audio fingerprint search technology, and more particularly, to a system and method for searching audio fingerprint by index information to improve recognition performance and to increase a search speed by indexing audio fingerprints, searching a predetermined audio fingerprint based on the indexing, and verifying the searched audio fingerprint.

[2] This work was supported by the Information Technology (IT) Research and Development Program of MIC (the Korean Ministry of Information and Communication ) / IITA (the Korean Institute for Information Technology Advancement) [2007-S-017-01, "Development of user-centric contents protection and distribution technology"].

[3]

Background Art

[4] The object of an audio fingerprint system is to recognize a predetermined audio by receiving an audio signal and searching a corresponding audio through a previously built audio fingerprint database. According to application fields, the audio fingerprint system has been used to broadcasting monitor, CF recognition, and file filtering. In order to effectively use the audio fingerprint system in the application fields, a high recognition rate and a fast search speed are required even under various distortions. Particularly, in order to filter files in P2P or UCC fields, it is required to quickly and accurately search an audio fingerprint data formed of several hundred thousand audio files each having own copyrights. The recognition speed is one of the most important factors for real-time processing in broadcasting monitoring and file filtering fields that operate based on the large capacity audio fingerprint database.

[5] Furthermore, it is also required for the audio fingerprint system to have a high recognition performance although audio data is deformed through re-sampling, filtering, equalization, and compression as well as the fast recognition speed according to the application fields of the audio finger print system.

[6] A search method according to the related art was introduced in Korean Patent Publication No. 2003-7001489 entitled "Method for search in audio database". In the method for search in audio database, a landmark and a fingerprint are extracted, and predetermined audio data is searched using a corresponding relation of a land mark and a fingerprint. In the method, a land mark is calculated beside a fingerprint, the calculated land mark is stored as an index, and a candidate list between a landmark and a music ID using a fingerprint in a landmark position. Then, the audio is recognized based on a linear relation thereof. However, the characteristics thereof were not considered although the audio signal was searched based on the fingerprint in the method. Also, the method needed a landmark to recognize a predetermined audio as a supplementary feature.

[7] An audio search system according to the related art was introduced in Korea Patent

Publication No. 2007-0031765 entitled "Fingerprint producing method and audio fingerprinting system based on normalized spectral subband centroids". The fingerprint producing method and audio fingerprinting system generate a fingerprint based on a normalized spectrum subband centroid and searches a predetermined audio by comparing distances of fingerprints. The fingerprint producing method and audio fingerprinting system did not consider the characteristics of a fingerprint for audio search although the fingerprint producing method and audio fingerprinting system had better recognition performance than MFCC and Tonality of typical fingerprints of MP3, equalization, and random start.

[8]

Disclosure of Invention Technical Problem

[9] Accordingly, the present invention is directed to a system and method for searching audio fingerprint by index information which substantially obviates one or more problems due to limitations and disadvantages of the related art.

[10] It is an object of the present invention to provide a system and method for searching audio fingerprint using index information to improve audio recognition performance and to increase a search speed by generating an index using the statistical characteristics of audio fingerprint feature information and searching a predetermined audio using the generated index.

[H]

Technical Solution

[12] To achieve these objects and other advantages and in accordance with the purpose of the invention, as embodied and broadly described herein, there is provided a system for searching an audio fingerprint including: a DB group for generating an index based on statistical characteristics of an audio fingerprint for an audio file and consecutively matching the index, the audio fingerprint, and music information; and an audio fingerprint searching apparatus for generating a new index based on statistical characteristic of an audio fingerprint for a new input audio file and searching corresponding music information for the new input audio file by searching the new index from the DB group.

[13] In accordance with another purpose of the invention, there is provided a method for searching an audio fingerprint using index information, including the steps of: a) generating an index based on statistical characteristics of an audio fingerprint for an audio file, and preparing a DB group for storing position information that consecutively matches the generated index, audio fingerprint, and music information; b) generating an index based on statistical characteristics of an audio fingerprint for a new input audio file; and c) searching corresponding music information for the new input audio file by searching the generated index, which is generated at the step b), from the DB group.

[14]

Advantageous Effects

[15] The system and method for searching audio fingerprint by index information according to the present invention generates an index using the statistical characteristics of an audio fingerprint and searches audio fingerprint based on the generated index. Therefore, the system and method for searching audio fingerprint by index information according to the present invention can sustain a fast search time and can be applied to the filtering and monitoring of files in a large capacity database. Furthermore, the system and method for searching audio fingerprint by index information creates candidate indexes that include an index bit of a variable position in order to compensate distortion because the recognition rate is abruptly degraded due to the distortion if the index is directly used to search without the compensation. Therefore, the system and method for searching audio fingerprint by index information can improve the recognition rate by correcting error that may be generated due to the bit index.

[16]

Brief Description of the Drawings

[17] The accompanying drawings, which are included to provide a further understanding of the invention, are incorporated in and constitute a part of this application, illustrate embodiments of the invention and together with the description serve to explain the principle of the invention. In the drawings:

[18] FIG. 1 is a block diagram illustrating a system for searching audio fingerprint according to an embodiment of the present invention;

[19] FIG. 2 is a block diagram illustrating an index processor according to an embodiment of the present invention;

[20] FIG. 3 is a diagram illustrating a relation in DB files used in a system for searching an audio fingerprint according to an embodiment of the present invention; [21] FIG. 4 is a diagram illustrating probability distribution used for generating a fingerprint index;

[22] FIG. 5 is a diagram illustrating a procedure of generating a fingerprint index using a fingerprint extracted from an audio search process and searching a predetermined audio based on the generated fingerprint index;

[23] FIG. 6 is a diagram illustrating a procedure of generating a candidate index in an audio search process; and

[24] FIG. 7 is a diagram illustrating a procedure of searching a final result using the candidate index generated at an audio search process.

[25]

Best Mode for Carrying Out the Invention

[26] Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings.

[27] FIG. 1 is a block diagram illustrating a system for searching audio fingerprint according to an embodiment of the present invention.

[28] Referring to FIG. 1, the system for searching audio fingerprint according to the present embodiment includes an audio fingerprint search apparatus 1. The audio fingerprint search apparatus 1 includes a fingerprint extracting unit 11 for extracting an audio fingerprint for an audio file, a candidate index searching unit 12 for generating candidate indexes in consideration of a variable position by sorting values of the extracted fingerpting in an ascending order of absolute values of differences between the extracted audio fingerprint value and a mean value that is used when an index is generated, a fingerprint matching unit 13 for matching an audio fingerprint to the extracted audio fingerprint corresponding to a candidate index, and a result verifying unit 14 for verifying that a search result is corresponding music information if a distance between audio fingerprints is in a predetermined value range.

[29] Since an index is used to search, the system for searching a fingerprint further includes a DB group 2 for storing audio fingerprints with corresponding indexes that are matched with the audio fingerprints. In order to operate the audio fingerprint search apparatus 1 according to the present invention, it is required to build a related database first. Therefore, it is preferable to form the DB group 2 to have a fingerprint DB 21, a music-information DB 22, and a fingerprint index DB 23. Also, in order to match and store audio fingerprints with corresponding indexes, it is required to perform a preprocess for an audio file. It is preferable to further includes an index processor 3 for receiving audio files and music ID information and matching audio fingerprints to corresponding indexes in the preprocess.

[30] As described above, the system for searching audio fingerprint according to the present embodiment is divided into a DB generating area for generating fingerprint indexes and building a database thereof and a DB searching area for searching through indexing. That is, related information is stored in the fingerprint DB 21, the music information DB 22, and the fingerprint index DB 23 in a DB generating step. Fingerprint extraction, candidate search and match through indexing, and result verification are performed in a step of searching based on an index. Although the system for searching audio fingerprint according to the present embodiment is divided into two areas, it is obvious to those skilled in the art that the two areas can be performed in one area.

[31] FIG. 2 is a block diagram illustrating an index processor according to an embodiment of the present invention.

[32] Referring to FIG. 2, the index processor 3 includes a fingerprint extractor 31, a fingerprint statistical analyzer 32, a fingerprint binarizor 33, and a fingerprint indexer 34.

[33] The fingerprint extractor 31 extracts an audio fingerprint from an input audio file using a fingerprint extraction algorithm, and the fingerprint statistical analyzer 32 analyzes statistical characteristics of an audio fingerprint and calculates a probability distribution. That is, the fingerprint statistical analyzer 32 approximates the extracted audio fingerprint to a probability model by calculating a mean and audio fingerprints and fingerprint distribution. Then, the fingerprint binarizor 33 binarizes the fingerprint to have probabilistic identical distribution by analyzing the statistical characteristics of audio fingerprint, and the fingerprint indexer 34 generates an index.

[34] Here, the fingerprint extraction algorithm may be one of zero crossing rate (ZCR), energy difference, spectral flatness, mel frequency cepstral coefficients (MFCC), and frequency centroide.

[35] FIG. 3 is a diagram illustrating a relation with the structure of a DB file used in a system for searching an audio fingerprint according to an embodiment of the present invention.

[36] Referring to FIG. 3, the DB group 2 according to the present embodiment includes the fingerprint DB 21, the music information DB 22, and the fingerprint index DB 23.

[37] The fingerprint DB 21 stores fingerprint audio values and position information in music. The music information DB 22 stores information about music ID and the number of fingerprints. The fingerprint index DB 23 stores information about position in the fingerprint DB 21 according to the binarized fingerprint value.

[38] When the system for searching audio fingerprint according to the present embodiment receives a request of searching a predetermined audio file after each of the DBs stores corresponding information, the system generates candidate indexes through the indexing step shown in FIG. 2, searches the generated candidate index from the fingerprint index DB 23, and detects fingerprint position information of the candidate index. Then, the system detects fingerprint information stored in the fingerprint DB 21 corresponding to the fingerprint position information and outputs the music information stored in the music information DB 22 corresponding to the position information of music.

[39] FIG. 4 is graphs illustrating probability distribution used for generating a fingerprint index.

[40] Referring to FIG. 4, the graphs show histogram distribution of normalized frequency centroid values. The graphs clearly show the mean value is close to 0.

[41] FIG. 5 is a diagram illustrating a procedure of generating a fingerprint index using a fingerprint extracted from an audio search process and searching a predetermined audio based on the generated fingerprint index.

[42] Like in FIG. 3, three DB files are prepared from an audio file. The audio fingerprint search apparatus 1 performs a search service using the three DB files.

[43] Referring to FIG. 5, when a predetermined audio file inputs, a step of extracting an audio fingerprint, a step of calculating a candidate fingerprint based on an index obtained from the extracted fingerprint, a step of matching the extracted audio fingerprint to an audio fingerprint corresponding to the calculated candidate fingerprint index , and a step of verifying a search result using the matching result are sequentially performed as the same method used for generating a DB from the audio file. These steps will be described as follows.

[44] If an audio file is inputted, an index is generated by extracting an audio fingerprint using the same method. That is, the fingerprint extractor 11 extracts an audio fingerprint from input audio files using a fingerprint extraction algorithm, and the fingerprint statistical analyzer 32 analyzes the statistical characteristics of the audio fingerprint and calculates probability distribution having the probabilistically identical distribution. That is, the fingerprint statistical analyzer 32 approximates a fingerprint to a probabilistic model by calculating the mean and distribution of the audio fingerprints. Then, the fingerprint binarizor 33 analyzes the statistical characteristics of the audio fingerprint and binarizes the audio fingerprint to have the probabilistically identical distribution. Then, the fingerprint indexer 34 generates an index.

[45] In order to obtain a candidate fingerprint value for audio search, information about a position in the fingerprint DB 21, which has a corresponding index value of the fingerprint index DB 23, is obtained. Meanwhile, if audio is distorted, a fingerprint extracted therefrom may be also distorted. Accordingly, the index value of a fingerprint may change. The index value may vary due to noise, equalization, compression, analog-digital conversion, and digital- analog conversion. Candidate indexes are generated for the index value variation.

[46] After generating the candidate index, the fingerprint index DB 23 obtains the information about a position having the corresponding candidate index value in the fin- gerprint DB 21. Then, the result of searching corresponding music information through the steps of matching and verifying a fingerprint.

[47] FIG. 6 is a diagram illustrating a procedure of generating a candidate index in an audio search process.

[48] As shown in FIG. 6, if N dimensional fingerprint values are present, the N dimensional fingerprint values are arranged in an ascending order based on the absolute values of differences with a mean value used for generating an index. A threshold value is decided according to the probability distribution shape, a variable position is decided, and candidate indexes are generated in consideration of the variable position.

[49] FIG. 7 is a diagram illustrating a step of searching a final result using the generated candidate index in the audio search step.

[50] As shown in FIG. 7, a fingerprint value is called, which is matched with the generated candidate index value, and a distance between a target audio fingerprint to search and a position having the called fingerprint value is calculated. Then, the calculated distance is compared with the predetermined threshold value. If the minimum value is smaller than the threshold value, one result is stored. In order to provide the high reliable result, the above mentioned steps are repeatedly performed at a fingerprint in a different position. Then, the final result is outputted through verifying the result.

[51] In the present embodiment, it is assumed that an extracted audio fingerprint has a floating-point real number value and that the estimated probabilistic model of a fingerprint has a bell shaped distribution which has a mean value of '0' as shown in FIG. 4. Here, N dimensional fingerprint can be expressed N binary numbers using the mean value of '0' as shown in Equation 1.

[52] [Equation 1]

[54] The fingerprint expressed in a binary number is converted to a decimal number through Equation 2. The decimal fingerprint number is used as an index for a database. [55] [Equation 2]

[56]

Index/^"i V = ∑ AFB J m] ■ 2 ^N"ra

[57] For example, in case of N is 16, a 16 dimensional audio fingerprint is expressed as one value between 0 to 65535, and this value is used as an index in a database.

[58] The audio fingerprint system according to the present embodiment generates three database files for audio search. As shown in FIG. 3, the three database files are formed as the fingerprint DB 21, the music information DB 22, and the fingerprint index DB 23.

[59] The fingerprint DB 21 stores extracted fingerprint values. That is, the fingerprint DB

21 stores the extracted fingerprint value as it is. The music information DB 22 stores information about music from which a fingerprint is extracted based on information provided when a fingerprint is generated. For example, the music information DB 22 may store various information such as a music ID, copyright information, a length of a fingerprint. The fingerprint index DB 23 transforms a fingerprint to an index through Equation 1 and Equation 2 and stores the fingerprint values, which are indexes, according to the position information in the fingerprint DB 21.

[60] For example, in case of 16-dimensional fingerprint, fingerprints are sequentially stored with position information as shown in FIG. 3. The music information and the fingerprint information are also stored with them. The fingerprint index is used for audio search by being stored with the position information of a fingerprint having a corresponding index value shown in FIG. 3.

[61] After preparing the three DB files from an audio file, the audio fingerprint system performs a search service using the prepared DB files. That is, if a predetermined audio file is inputted, a step of extracting an audio fingerprint using the same method used for generating the DB from the audio file, a step of calculating candidate fingerprints by calculating indexes from extracted fingerprints, a step of matching candidate fingerprints, and a step of verifying using a matching result are sequentially performed. Such steps will be described in more detail as follows.

[62] When an audio file is inputted, an audio fingerprint is extracted using the audio fingerprint extraction method, and an index is generated based on the extracted audio fingerprint.

[63] In order to obtain a candidate fingerprint value for audio search, information about a position in the fingerprint DB 21, which has a corresponding index value of the fingerprint index DB 23, is obtained. Meanwhile, if audio is distorted, a fingerprint extracted therefrom may be also distorted. Accordingly, the index value of a fingerprint may change. The index value may vary due to noise, equalization, compression, analog-digital conversion, and digital- analog conversion. Candidate indexes are generated for the index value variation.

[64] After generating the candidate indexes, position information in the fingerprint DB

21, which has a corresponding candidate index value in the fingerprint index DB 23, is obtained corresponding to the candidate indexes. Then, a search result of corresponding music information is outputted after the fingerprint matching and verifying steps.

[65] Hereinafter, the step of generating a candidate index will described in more detail. In case of N-dimensional fingerprint, an N-dimensional value is arranged in an order of closest distances from a mean value of a probabilistic distribution for generating an index, for example a mean value of '0' in the present embodiment. Here, positions having the large probability of changing according to the probabilistic distribution can be sequentially calculated. Positions in a predetermined distance range through probabilistic distribution can be selected. Also, a predetermined number of positions can be selected without any condition.

[66] [Equation 3]

[67] sort_ascend{| F— meanJ}

[68] If a threshold value having a bell shaped probabilistic distribution is decided as a constant number in a previously used audio fingerprint, information about positions changeable according to the fingerprint can be obtained. After the position is decided, an index is generated using the index generation method used in the step of extracting a fingerprint. In addition, all possible indexes are generated corresponding to variable positions.

[69] For example, if a 4-dimensional audio finger value is (-0.2, 0.1, 0.4, 0.2), if a value deciding an index is 0, and if the 2^nd position has the large probability to change, an index thereof is 0111 and a candidate index is 0011 because the 2^nd position may change.

[70] After calculating the candidate indexes as described above, candidate fingerprints are obtained from a corresponding index with reference to the position information in the fingerprint DB 21 and the obtained candidate fingerprints are arranged by comparing distances to a target audio fingerprint to search according to the position information.

[71] Herein, redundancy is removed based on values corresponding to reference positions as a reference, and distances from K predetermined audio fingerprints having a predetermined length to a fingerprint value of a fingerprint DB 21 are calculated. For example, a Euclidian distance is calculated, and the calculated Euclidian distance is compared with a threshold value. If the calculated is smaller than the threshold value, music information is searched in the music information DB 22 and the search result is outputted. If not, a basic information value indicating that music is not searched is outputted as a result. The above mentioned steps were well described with reference to FIG. 7. The result value is formed of Music ID denoting music information in a data base, Position denoting a temporal position, and Distance denoting distance difference that is reliability as shown in Equation 4.

[72] [Equation 4]

[73] Rf ny = { MusicID[nj, Posit ion[nj,Di stance/^" n]}

[74] When the value of Music ID is smaller than a threshold value, the value of Music ID is calculated using position information in the fingerprint index DB 23. Since the music information DB 22 stores the number of fingerprints of each music, the value of a position is larger than the sum of the fingerprint numbers up to (m-l)^th music if the m th Music ID is a result. Also, the value of a position is smaller than the sum of the fingerprint numbers of music up to the (m+l)"¹ music. Using this fact, the value of Music ID is calculated.

[75] [Equation 5]

[76] m-l m+1

∑ feat_num [k] < pos i t i on/^{^}ny < ^ feat_num [kj k= l k = l

[77] A general system may perform the searching step several times in order to improve the reliability of search. After calculating a candidate index at a predetermined position, a candidate index may be searched again at another position, and a step of searching a candidate fingerprint is performed repeatedly so as to obtain a result.

[78] Such a result is decided based on a parameter value that is selected by a system, and the search results are stored as many as the number of times of performing the searching step. The stored results are outputted as a final searching result after the verification step.

[79] In the verification step, fingerprints extracted from a predetermined audio signal to search are consecutive values in a time domain. That is, fingerprints are sequentially extracted in time. Similarly, the fingerprints are sequentially extracted and stored in time at the generated fingerprint DB 21.

[80] That is, a result of searching using a fingerprint at a predetermined position and a result of searching using a fingerprint at a next position have the temporally identical distance difference. Based on this fact, the searching result is verified using Equation 6.

[81] [Equation 6]

^[82] i f Musi cID/n + p7 = Musi cID/ny, p - 1 < Pos i t ion [n + p] - Pos i t ion [n] < p + 1

[83] That is, if searching results of p positions are the same, the difference of position information must be larger than p-1 and smaller than p+1. Based on this fact, repeatedly obtained results are verified. If the condition is satisfied, the results are outputted as a final result.

[84] A test database is generated for 27,000 audio files each having a length of 40 seconds for verification. Then, 100 audio files, which are compressed to 32kbps-MP3 audio files each having a length of 20 seconds, are searched from the test database. A fingerprint described in the present embodiment is extracted for 16-dimensions and the extracted fingerprint is used. In order to compare distances, 52 fingerprints are used. Also, the search step is performed five times for verification. In order to compare performances, a sequential search is performed under the same condition. In the sequential search, all fingerprints are searched in a DB and a result having a minimum value by comparing distances is determined as a final result.

[85] Table 1 [Table 1] [Table ]

[86] As shown in Table 1, the search speed of the index bases search according to the present embodiment is much faster than that of the sequential search. Also, the deterioration of the recognition rate for the bit index value can be overcome by adjusting a parameter value that decides a candidate index although the total search time extends little bit.

[87] That is, the recognition rate can be improved from 87% to 96% although the total search time extends as long as about 10 seconds. The search time can be reduced up to 1/9 of the total search time of the sequential search.

[88] The system and method for searching audio fingerprint by index information according to the present embodiment can be applied to the filtering and monitoring of files in a large capacity database based on the fast search time and the high recognition rate. Particularly, the system and method for searching audio fingerprint by index information according to the present embodiment can be applied to the file filtering to solve the copyright problem in a user created content (UCC) field or a P2P field. [89] It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention. Thus, it is intended that the present invention covers the modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents.

Claims

[1] A system for searching an audio fingerprint comprising: a DB group for generating an index based on statistical characteristics of an audio fingerprint for an audio file and consecutively matching the index, the audio fingerprint, and music information; and an audio fingerprint searching apparatus for generating a new index based on statistical characteristic of an audio fingerprint for a new input audio file and searching corresponding music information for the new input audio file by searching the new index from the DB group.

[2] The system of claim 1, wherein the DB group includes: a fingerprint DB for storing audio fingerprints for the audio file and position information of the music information; a music information DB for storing music ID for the music information and information about the number of fingerprints; and a fingerprint index DB for storing information about audio fingerprint position in the fingerprint DB corresponding to the index.

[3] The system of claim 2, further comprising an index processor for transferring audio fingerprints and music information, which are extracted by extracting an audio fingerprint from an audio file, to a corresponding DB.

[4] The system of claim 3, wherein the index processor includes: a fingerprint extractor for extracting an audio fingerprint using a fingerprint extraction algorithm; a fingerprint statistic analyzer for approximating a fingerprint to a probabilistic model by obtaining a mean and distribution of audio fingerprints for the extracted audio fingerprint; a fingerprint binarizor for by analyzing statistical characteristics of the audio fingerprint and performing binarization to have a probabilistically identical distribution; and a fingerprint indexer for matching the binarization result to an index.

[5] The system of claim 4, wherein the fingerprint extraction algorithm is one of

Zero Crossing Rate (ZCR), Energy Difference, Spectral flatness, Mel Frequency Cepstral Coefficients (MFCC), and Frequency Centroids.

[6] The system of claim 4, wherein the binarization is performed based on a mean value among the statistical characteristics.

[7] The system of claim 1, wherein the audio fingerprint searching apparatus includes: a fingerprint extractor for extracting an audio fingerprint for the new audio file; a candidate index searching unit for generating a candidate index by sorting values of the extracted audio fingerprint in an ascending order of absolute values of differences between the extracted audio fingerprint value and a mean value, which is used for generating the new index, and in consideration of a variable position; a fingerprint matching unit for matching an audio fingerprint to the extracted audio fingerprint corresponding to the candidate index; and a result verifying unit for measuring a distance between the audio fingerprints and verifying a result through time information if the measured distance is in a predetermined value range.

[8] The system of claim 7, wherein the variable position is decided by a threshold value setting in a probabilistic distribution shape.

[9] A method for searching an audio fingerprint using index information, comprising the steps of: a) generating an index based on statistical characteristics of an audio fingerprint for an audio file, and preparing a DB group for storing position information that consecutively matches the generated index, audio fingerprint, and music information; b) generating an index based on statistical characteristics of an audio fingerprint for a new input audio file; and c) searching corresponding music information for the new input audio file by searching the generated index, which is generated at the step b), from the DB group.

[10] The method of claim 9, wherein the index generation in the steps a) and b) includes the steps of: extracting an audio fingerprint using a fingerprint extraction algorithm; approximating a fingerprint to a probabilistic model by calculating a mean value and distribution of an audio fingerprint for the extracted audio fingerprint; and generating an index by analyzing the statistical characteristics of the audio fingerprint and performing binarization to have a probabilistically identical distribution.

[11] The method of claim 9, wherein the step a) includes the steps of: storing an audio fingerprint for the audio file and position information for the music information at a fingerprint DB; storing an Music ID that is an unique ID for the music information and information including the number of fingerprints in a music information DB; and storing information about audio fingerprint position in a fingerprint DB corresponding to the index at a fingerprint index DB.

[12] The method of claim 9, wherein the step c) includes the steps of: generating a candidate index by sorting values of the extracted audio fingerprint in an ascending order of absolute values of differences between the extracted audio fingerprint value and a mean value used for generating the index in the step b) and in consideration of a variable position; matching an audio fingerprint corresponding to the candidate index to the extracted audio fingerprint; and verifying a result through time information by measuring a distance between the audio fingerprints and the measured distance is in a predetermined value range.

[13] The method of claim 12, wherein a candidate index is generated after absolute values of differences with a mean value, which is used for generating the index, are sorted in an ascending order and a bit value of a corresponding position changes by deciding a position of a dimension close to a mean value according to a threshold value.

[14] The method of claim 13, wherein Euclidian distances between fingerprint position information calculated from the candidate index and a fingerprint obtained from the new input audio file are calculated as many as a predetermined number with reference to a fingerprint DB, and music information having a minimum distance is searched.

[15] The method of claim 14, wherein music information of a fingerprint position having a minimum distance is outputted as a result if the measured distance is in a threshold value range in the candidate index.

[16] The method of claim 15, wherein the music information search is outputted as a final result using fingerprint of other positions if a distance difference between fingerprints in the new input audio file is identical to position information difference of a result.