Patent Grant
7277766
The present application is related to the U.S. patent application entitled “METHOD FOR CREATING A DATABASE FOR COMPARING MUSIC ATTRIBUTES”, Ser. No. 09/533,045, filed on Mar. 22, 2000, and assigned to the Assignee of the present invention is hereby incorporated by reference in its entirety.
The present invention relates to analyzing audio files and more particularly to presenting a playlist based upon listener preferences and audio file content.
The Internet connects thousands of computers world wide into a vast network using well-known protocols, for example, Transmission Control Protocol (TCP)/Internet Protocol (IP). Information on the Internet is stored world wide as computer files, mostly written in the Hypertext Mark Up Language (“HTML”). The collection of all such publicly available computer files is known as the World Wide Web (WWW).
The WWW is a multimedia-enabled hypertext system used for navigating the Internet and is made up of hundreds of thousands of web pages with audio, images, text and video files. Each web page can have connections to other pages, which may be located on any computer connected to the Internet.
A typical Internet user uses a client program called a “Web Browser” to connect to the Internet. A user can connect to the Internet via a proprietary network, such as America Online or CompuServe, or via an Internet Service Provider, e.g., Earthlink.
A Web Browser may run on any computer connected to the Internet. Currently, various browsers are available of which two prominent browsers are Netscape Navigator and Microsoft Internet Explorer. The Web Browser receives and sends requests to a web server and acquires information from the WWW. A web server is a program that, upon receipt of a request, sends the requested data to the requesting user.
A standard naming convention known as Uniform Resource Locator (“URL”) has been adopted to represent hypermedia links and links to network services. Most files or services can be represented with a URL. URLs enable Web Browsers to go directly to any file held on any WWW server.
Information from the WWW is accessed using well-known protocols, including the Hypertext Transport Protocol (“HTTP”), the Wide Area Information Service (“WAIS”) and the File Transport Protocol (“FTP”), over TCP/IP protocol. The transfer format for standard WWW pages is Hypertext Transfer Protocol (HTTP).
The advent and progress of the Internet has changed the way consumers buy or listen to music. Consumers today can download digital music via the Internet using MP3 or SDMI technology, with a click of a mouse. Audio delivery techniques have also made it easy to stream audio from a website to a consumer, upon demand. A typical music listener can download audio files from the WWW, store the audio files, and listen to music.
Currently music can be stored in various file formats. Generally there are two types of file formats: (1) self-describing formats, where device parameters and encoding are made explicit in a header, and (2) headerless formats, where device parameters and encoding are fixed.
The header of self-describing formats contain parameters of a sampling device and may also include other information (e.g. a human-readable description of sound, or a copyright notice etc.). Some examples of popular self describing formats are provided below:
Headerless formats define single encoding and usually allow no variation in device parameters (except sometimes for sampling rates). The following are a few examples of Headerless formats:
Although music listeners can store audio files, conventional music search techniques do not allow a music listener to search for music based upon audio file content. Conventional systems also do not allow a music listener to generate play lists based upon music listener preferences and/or audio file content.
Hence what is needed is a method and system that can analyze audio file content and produce a play list based upon preferences defined by a music listener.
The present invention solves the foregoing drawbacks by providing a method and system for analyzing audio files. Plural audio file feature vector values based on an audio file's content are determined and the audio file feature vectors are stored in a database, that also stores other pre-computed audio file features. The process determines if the audio files feature vectors match the stored audio file vectors. The process also associates a plurality of known attributes to the audio file.
The present invention includes a system for analyzing audio files that includes a playlist generates that determines a plurality of audio file vectors based upon an audio file's content; and a signature comparator between the playlist generator and a database, wherein the database stores a plurality of audio file vector values of plural music samples. The signature comparator compares input audio samples with previously stored audio samples in the database. Also provided is a user interface that allows a music listener to input search request for searching music based upon attributes that define music content.
In another aspect the present invention includes a method for determining audio signatures for input audio samples. The process extracts plural features representing the input audio samples, wherein the features are extracted by Fourier transform analysis. The process also identifies a set of representative points based upon the plural features, and determines a code book of plural elements for mapping the representative points to the elements of the code book.
In yet another aspect, the present invention includes a method for comparing input audio signatures with pre-computed stored audio signatures. The process determines a query signature based upon the input audio signature and divides the query signature into a string of characters; and compares the string of characters to stored pre-computed audio signatures.
In yet another aspect, the present invention divides an input audio sample into bins and determines a plurality of features describing the bins. Thereafter, the process determines a univariate signal based upon the plural features and computes an audio signature based upon the univariate signal.
One advantage of the foregoing aspects of the present invention is that unique audio signatures may be assigned to audio files. Also various attributes may be tagged to audio files. The present invention can generate a customized playlist for a user based upon audio file content and the attached attributes. Hence making the music searching experience easy and customized.
This brief summary has been provided so that the nature of the invention may be understood quickly. A more complete understanding of the invention can be obtained by reference to the following detailed description of the preferred embodiments thereof in connection with the attached drawings.
The use of similar reference numerals in different figures indicates similar or identical items.
Computer 10 includes a computer-readable memory medium such as a rotating disk 15 for storing readable data. Besides other programs, disk 15 can store application programs including web browsers by which computer 10 connects to the Internet and the systems according to the present invention as described below.
Computer 10 can also access a computer-readable floppy disk storing data files, application program files, and computer executable process steps embodying the present invention or the like via a floppy disk drive 16. A CD-ROM interface (not shown) may also be provided with computer 10 to access application program files, audio files and data files stored on a CD-ROM.
A modem, an integrated services digital network (ISDN) connection, or the like also provides computer 10 with an Internet connection 12 to the World Wide Web (WWW). The Internet connection 12 allows computer 10 to download data files, audio files, application program files and computer-executable process steps embodying the present invention.
Computer 10 is also provided with external audio speakers 17A and 17B to assist a listener to listen to music either on-line, downloaded from the Internet or off-line using a CD (not shown). It is noteworthy that a listener may use headphones instead of audio speakers 17A and 17B to listen to music.
As described above, disk 15 stores operating system program files, application program files, web browsers, and other files. Some of these files are stored on disk 15 using an installation program. For example, CPU 201 executes computer-executable process steps of an installation program so that CPU 201 can properly execute application programs.
A random access main memory (“RAM”) 206 also interfaces to computer bus 208 to provide CPU 201 with access to memory storage. When executing stored computer-executable process steps from disk 15 (or other storage media such as floppy disk 16 or WWW connection 12), CPU 201 stores and executes the process steps out of RAM 206.
Read only memory (“ROM”) 207 is provided to store invariant instruction sequences such as start-up instruction sequences or basic input/output operating system (BIOS) sequences for operation of keyboard 13.
The present invention is not limited to the computer architecture described above. Systems comparable to Computer 10, for example, Portable devices hand held computing devices that can be connected to the Internet may also be used to implement the present inventive techniques.
It is noteworthy that the invention is not limited to a particular number of computers. Any number of computers that can be connected to the Internet 302 or to any other computer network may be used to implement the present inventive techniques.
A playlist generator 400 accesses audio files stored on rotating disk 15. Playlist generator 400 is an application program that can be located on remote server 304 or on a music listener's computer 10. Playlist generator 400 scans and analyzes audio files stored on rotating disk 15 and computes audio signatures that uniquely and compactly identify the content of the audio file. An audio signature is computed only once for each file and stored in a local database. The computed audio signatures that compares the analyzed audio files with previously analyzed audio data is stored in a central database 401. Central database 401 stores a plurality of feature vector values as discussed below. Central database 401 also includes data similar to the data stored in a production database that is described in U.S. patent application Ser. No. 09/533,045, entitled, “Method for Creating a Database for Comparing Music Attributes”, incorporated herein by reference in its entirety. Customized play lists are generated after audio file content is compared to the data stored in central database 401, as described below.
A user interface 400A is also provided that allows a music listener to input preferences for generating play lists. User interface 400A may be a separate component or integrated with playlist generator 400. One such user interface is described in U.S. patent application Ser. No. 09/533,045, entitled Method for Creating a Database for Comparing Music Attributes, incorporated herein by reference in its entirety.
The Signature Comparator 400C receives user requests from UI 400A and obtains music or a list of music based upon user preferences, as described below. It is noteworthy that audio file analyzer 400B and signature comparator 400C may be integrated into a single module to determine audio signature and implement the various aspects of the present invention.
In step S601 audio analyzer 400B receives audio files (“input audio files”). Audio files may be stored on a user's disk 15 or at webserver 304 connected to the Internet or on any private network. Audio analyzer 400B may seek audio files based upon user input in UI 400A or may receive audio files from a designated source.
In step S602, audio analyzer 400B analyzes input audio file content and computes a set of parameters (“audio file vectors”). Audio file vectors can uniquely identify audio files and can also be used to assign a unique “audio signature” for an input audio file. Details of determining audio signatures are described below. These audio signatures need be computed only once for a given audio file and can thereafter be stored either locally or remotely and referred to as necessary.
In step S603, input audio file vector values for specific bins and the audio signature are stored in a database. One such database is a central database 401. Audio file signatures are sets or vectors of values based upon the audio segments as shown in
In step S701, playlist generator 400 scans an audio file(s) stored on disk 15. Audio files may be pulled by playlist generator 400 from disk 15, or a music listener may send audio files to playlist generator 400 via the Internet 302 or a private network.
In step S702, playlist generator 400 determines the audio signature of input audio file(s). An audio signature may be determined by the process described below.
In step S703, playlist generator 400 transfers the audio signature and audio file vector values (Vt1 . . . Vtk) to Signature Comparator 400C. Playlist generator 400 also commands the Signature Comparator 400C to compare the audio signature and feature vectors values determined in step S702 with historical audio signatures and audio file vector values stored in central database 401. Thereafter, the Signature Comparator 400C compares input audio signature and audio file vector values with historical audio signatures and audio file vector values, stored in database 401.
In step S704, Signature Comparator 400C determines whether the input audio signature and audio file vectors values match stored audio signatures and vector values. If the input audio file do not match with any stored audio data (audio signature and feature vector), then in step S705 input audio file vector values are is stored in central database 401.
If the input audio file signature matches with stored audio signatures and vector values, then in step S706, the process determines if the stored entries are confirmed or provisional. A confirmed entry in database 401 is an entry that has been ratified by multiple sources.
A provisional entry in database 401 is one that is not confirmed.
If the input audio file matches a confirmed entry, then in step S707 other feature values stored in database 401 are associated with the audio file. Examples of such feature vectors are provided in U.S. patent application Ser. No. 09/533,045, entitled Method for Creating a Database for Comparing Music Attributes, filed on Mar. 22, 2000, assigned to the assignee herein, and incorporated by reference in its entirety. Associating a plurality of feature vectors allows a listener to search for music based upon content. For example, feature values associated with a given audio file may indicate that it is Rock music with a slow tempo and a smooth, female singer accompanied by a band featuring a prominent saxophone.
Some of the features that can be associated with the audio files are:
(a) Emotional quality vector values that indicates whether an audio file content is Intense, Happy, Sad Mellow, Romantic, Heartbreaking, Aggressive or Upbeat.
(b) Vocal vector values that indicates whether the audio file content includes a Sexy voice, a Smooth voice, a Powerful voice, a Great voice, or a Soulful voice.
(c) Sound quality vector values that indicates whether the audio file content includes a strong beat, is simple, has a good groove, is fast, is speech like or emphasizes a melody.
(d) Situational quality vector values that indicate whether the audio file content is good for a workout, a shopping mall, a dinner party, a dance party, slow dancing or studying.
(e) Ensemble vector values indicating whether the audio file includes a female solo, male solo, female duet, male duet, mined duet, female group, male group or instrumental.
(f) Genre vector values that indicate whether the audio file content belongs to a plurality of genres including Alternative, Blues, County, Electronics/Dance, Folk, Gospel, Jazz, Latin, New Age, Rhythm and Blues (R and B), Soul, Rap, Hip-Hop, Reggae, Rock and others.
(g) Instrument vectors that indicates whether the audio file content includes an acoustic guitar, electric guitar, bass, drum, harmonica, organ, piano, synthesizer, horn or saxophone.
If the input audio file matches a provisional rating, then in step S708, the process converts the provisional rating to a confirmed rating.
In step S801, a music listener inputs a request for a playlist. UI 400A (
In step S802, playlist generator 400 scans audio files stored on user's disk 15. One methodology of analyzing audio files is provided above in
In step S803, Signature Comparator 400C searches for music based upon analyzed audio file data.
In step S804, playlist generator 400 compiles a playlist based upon user preferences and the compared data.
Determining Audio Signatures
Audio signatures identify audio files and are based upon the signal characteristics of a particular audio file sample. If the audio signatures of a group of audio files is known then a new audio file may be compared with the known group of audio signatures. Audio signature is a representation of an audio file that assists in comparing audio samples. Audio signature may be developed for any audio file and whenever two audio samples overlap, the audio signatures of the samples will also overlap. It is this property that assists comparison of audio samples.
In step S901, an audio sample is represented as a signal. The signal is then used to determine a set of parameters or features that describe the signal.
In step S 902, the determined features from step S901 are used to compute an audio signature for the audio sample.
In step S903, the audio signature determined in step S902 is compared pre-computed audio signatures stored in database 401.
The following describes different techniques under the present invention that may be used to implement the process steps of
Vector Quantization Methodology
Under this methodology, an audio file is transformed into a raw intensity signal and the transformed signal is used to compute a set of features or parameters. The features are computed from sequential (possibly overlapping) sets of raw intensity signal and transforms the intensity signal into a time series of feature values. This multivariate time series of feature vectors is then compressed into a 1-dimensional vector of elements. This compression is achieved using vector quantization of the multivariate features, so that each point in the feature space can be mapped onto an element in a finite code book. Thereafter, the computed 1-dimensional vector (1-d string) is matched against known values stored in a database using fast string matching algorithms to retrieve a match.
Pre-Processing
As discussed in step S901, pre-processing is used to convert an audio file whose signature is to be calculated, to a standard reference format. More specifically mono audio files of 16-bit samples at 22050 Hz sampling rate are used as a reference format. Standard commercially available software converters and decoders may be used for this purpose. More specifically Sox sound converting utility that can be downloaded from the Internet address www.spies.com/Sox/ and Xaudio mp3 decoder downloaded from a website located at www.xaudio.com may be used.
Feature Extraction
As discussed in step S901, certain features or parameters are extracted from an audio file signal. The features of this methodology are based on Short Time Fourier Transform (STFT) analysis. The STFT is a signal processing analysis technique that processes audio file signals in samples designated as bins or windows (
It is noteworthy that because overall level (volume) effects vary across different encodings of the same audio file and can not be assumed to be uniform, STFT features are normalized to eliminate level (volume) effects by dividing out the dc component of the STFT at each sample point before further processing
The following STFT-based features may be extracted in step S901:
Spectral Centroid is the balancing point of the spectrum (The spectrum is the representation of audio intensity over time. (
where N is the FFT window size and the Mi's are the frequency bins of the spectrum magnitude.
Spectral Rolloff like the spectral centroid is another measure of the shape of the spectrum. It is defined as:
R=r such that Σ1rMi=0.8*ΣrNMi
where N is the FFT window size, the M_i's are the frequency bins of the spectrum magnitude.
Spectral Flux is the 2-norm of the difference between the magnitude of the short time Fourier transform (STFT) spectrum evaluated at two successive analysis windows. Note that the signal is first normalized for energy, i.e. all segments on which the STFT is calculated are constrained to have equal energy (the same dc value of the STFT).
Peak ratio is the ratio of the magnitude of the highest peak in the magnitude spectrum to the average magnitude.
Subband energy vector is a vector of sub-band energies calculated by grouping STFT frequency bins to logarithmically spaced sub-bands. The STFT is normalized in energy.
Subband flux is the 2-norm of the difference between the subband energy vectors evaluated at two successively analyzed windows.
Subband Energy Ratios are formed by analyzing the energy component of the signal in each of a defined set of frequency bands. The ratios of the energy in these sub-bands provide a set of measures of the spectral composition of the signal. For example, if energies in 5 sub-bands are determined then there are 10 (5 choose 2) distinct ratios between the 5 sub-bands. These 10 numbers define a vector, which characterizes the spectral energy distribution. A related set of statistics that may be used is the logarithmic ratios of the sub-band energies. The logarithmic ratios makes the numbers more stable and robust.
Signature Calculation (Step S902,
Vector Quantization
Feature extraction in step S901 provides a time series of feature vectors. Vector Quantization (VQ) is a technique that maps a large set of vectors to a smaller representative indexed set of vectors (code words) designated as the code-book. VQ is used to compress vector information. VQ reduces multidimensional vectors comprising the computed features into a single value.
In step S101, a set of representative points are identified.
In step S102, the process defines a code-book and each of the representative points derived in step S101 above is mapped to an element in the code book, where each element corresponds to a unique ASCII character.
In step S103, the process defines a rule that can map any point in a feature space onto an element of the codebook. Cluster centers are assigned to the elements of the codebook. Points that are not cluster centers are mapped to a cluster center, and then assigned to an element of the codebook corresponding to that cluster center. To map an arbitrary point onto a cluster center Mahalonobis distance metric and a nearest neighbor algorithm is used. Every point is mapped onto the closest cluster center, using a Mahalonobis distance or a similar metric.
VQ provides a string of characters. These characters may be compressed by using a logarithmic run length compression scheme. In this scheme a run of the same character is compressed to a shorter run with length equal to the logarithm of the original run length. For example, the string of characters aaaabcccccccc will be compressed to aabccc (Log28=3 so the 8 c's are compressed to 3. The compressed string provides compact audio signatures that can be efficiently transmitted and compared. Also, logarithmic run length compression highlights the regions when the signal is changing.
Base Overshifting
The process described above depends upon STFT which in turn depends on the bin or window size of the signal. In particular, to compute STFT efficiently, the raw audio signal is partitioned into bins of a certain size (for example bins size 512 of sample points). However, where the bins should start and end affects the accuracy of the computed audio signature. For example, if a raw signal has 10,000 points and the signal is divided into bins of length 512, the computation could start at the beginning of every 512 points or the computation could start at the second, or tenth, or hundredth point etc. Since not every audio sample starts at the “beginning” of some canonical version the exact choice of bins or windows the signature extraction procedure should be robust to average out the arbitrariness in the computation.
The present solution provides a solution to the arbitrariness of bin selection by choosing features that are intrinsically robust to shifts in binning. Also, multiple signatures may be computed for each audio sample based on different bin shifts. For example, one signature is computed based on starting the first bin at the beginning of the audio file. A second signature is then computed based on bins starting at the Nth data point. A third signature might be computed based on the first bin starting at the N+Kth data point and so on. Plural number of shifted audio signatures is computed for each audio sample. Each audio signature is a short list of strings (typically of length 8 or 16) where each string corresponds to a slightly shifted bin analysis.
Audio signatures determined by the foregoing process allows signature comparison of new audio samples (query signature) with stored pre-computed audio signatures (base signature)(
Assume that a query signature is denoted as AABBCADAABC . . .
The above signature is broken into sub-strings of defined length (e.g. 11 characters). The sub-string of 11 characters is then compared to the stored base signature stored in central database 401.
Hierarchical matching may also be used to obtain the best match. In this process a sub-string of X characters is used and thereafter, a subset of the sub string comprising of X′ characters where X′ is less than X is used to match a query signature with the base algorithm. A specific confidence level or value may be assigned to a match for each sub-string. A match index value may be used to enhance the accuracy of matching where match index value is given by:
MI=⊂Wi*Ni
Where Wi is a weight (real number) for matches of length i and Ni is the number of substring matches of length i. Thus, Match Index (MI) is a weighted sum of the number of substring matches of different length substrings. Matches for longer substrings are less likely by chance and should thus be assigned a higher weight than matches of shorter length. Thus, the magnitude of the weight Wi is a measure of the confidence assigned to matches of a given substring length.
The foregoing process can be iterated to improve performance. For example, after a particular code-book is defined to match a small set of characters out of a large set, the process may define a second code-book for the subset of songs matched with the first codebook to expedite matching. Alternatively, many independent codebooks may be defined and use them in conjunction. The best matching songs will match across the most codebooks. The advantage of this approach is that errors in matching in one codebook will be compensated for by matches in another codebook.
The following specific values may be used for the foregoing process:
sampling_rate=22050 Hz
sample size (bits)=16
fft_size=512
window_size=512
code_book_size=14
overshifting=8
query substring length=14
query overlap=13
It is noteworthy that the foregoing values are merely to illustrate the foregoing process and is not to limit the invention.
Segmentation Technique
Another method for determining audio signatures according to the present invention involves the following 1) an audio sample is segmented into disjoint, contiguous regions in a robust and deterministic manner, 2) a robust description of the audio characteristics of each segment is determined. This description may be any parameterization of the audio file, or may consist of nonparametric samples, and 3) an audio signature is developed which reflects the segmentation structure of the original audio sample. One method to determine audio for segmenting audio files is described in an article “Multifeature Audio Segmentation for Browsing and Annotation”, by George Tzanetakis, one of the present inventors, and Perry Cook (“Tzanetakis and Cook”), published Oct. 17-20, 1999, in “IEEE Workshop on Applications of Signal Processing to Audio and Acoustics”, incorporated herein be reference in its entirety:
In step S1001, an input audio file's is plotted over time.
In step S1002, the process determines audio sample feature vectors or parameters. A set of parameters for Ia is computed for a set of short contiguous bins (or short overlapping bins). Hence, Ia is transformed into a set of parametric descriptions (over time), designated for the purpose of illustration as Pa1, Pa2, . . . , PaN, for N parameterizations. Pa1, Pa1, . . . , PaN, is a time series of feature vectors that describe the audio file intensity Ia. Examples of such parameters are described above under vector quantization methodology and also include the following:
(a) Zero Crossings: Number of time domain zero crossings, i.e, the number of times the signal intensity changes from positive to negative or negative to positive in the sample file.
(b) Root Mean Square (RMS) energy (intensity) within a given bin.
(c) Nth percentile frequency of the spectrum (e.g., N=25, 50, 75, etc)
The Nth percentile frequency of the spectrum of a specific bin is the largest frequency fk such that Sm(Ai)/Sum(Aj)<=N*100 where i=1,2, . . . , fk and j=1, 2, . . . , fM where fM is the highest sample frequency.
(h) Spectral energy correlation of contiguous bins (the correlation between bins i and i+N (where N is a positive integer representing the correlation lag computed for the spectral energy across frequencies).
The foregoing list of parameters is not exhaustive. Other mathematical variables may also be used to determine audio file vectors.
In step S1003, the process determines a univariate signal based upon the foregoing audio file parameters. Pa1, Pa2, . . . , PaN of Ia, are combined into a single, univariate signal dt. One method to compute dt is to compute the distance between successive time frames of the feature vector of parameterizations. Various distance metrics may be used as discussed in Tzanetakis and Cook. Mahalonobis distance is one such metric and is used between successive values of feature vectors Pa1, Pa2, . . . , PaN. Other distance metrics such as, Minkowski metrics, may also be used.
An example of determining dt is provided below. Assume that t1 and t2 are adjacent bins of Ia which have parameterization feature vectors V1 and V2 respectively.
Hence,
V1=[Pa11Pa21 . . . PaN1]
and
V2=[Pa11Pa21 . . . PaN1].
dt=∥V2−Vt∥ where ∥ ∥ indicates computation of the Mahalonobis distance between V1 and V2.
Therafter Δti is determined by ∥Vt−Vt−1∥ for all adjacent bins
After determining the univariate distance signal Δt, in step S1004, the process segments the audio sample. Regions of high change in the signal define segmentation. A derivative,
is computed and compared to a predefined threshold value. Points at which the derivative value is greater than the threshold value indicate sudden transitions in the audio signal and define segmentation points.
In step S1005, the audio signature for a sample audio file is determined, as illustrated by the following example.
For a given audio sample S, segment points are determined at t1, t2, . . . , tN using the procedure described above. A set of N−1 segments may then be constructed, having lengths L1, L2, . . . , LN−1 where Lk=tk+1−tk and tk is the time point (in the sample) of the kth segment. For each audio segment a set of robust statistics are computed. Numerous statistical values may be used to define and describe specific segments for example, spectral centroid; spectral rolloff; intensity variance, skewness, kurtosis; and sub-band energy ratios etc
Hence, determined a set of audio segments L1, L2, . . . , LM and a set of k robust statistics for each segment R11, R12, . . . , R1k, an audio signature, (“AS”) may be determined as follows:
AS=[L1 R11 R12 . . . R1k L2 R21 R22 . . . R2k . . . LM RM1 RM2 . . . RMk]
AS is a vector or string which concatenates the segment length and robust summary for each of the segments.
It is noteworthy that AS is definable for any audio sample of sufficient length to contain at least 1 segment, and if two different audio samples contain overlapping audio segments, then the audio signatures contain overlapping elements.
It is noteworthy that instead of segmenting an audio file into variable length segments as described above, the audio file is first segmented into fixed-length segments, and then robust statistics are calculated within each segment.
It is also noteworthy that any or all of these techniques described above may be combined in a multistage procedure.
One advantage of the foregoing aspects of the present invention is that unique audio signatures may be assigned to audio files. Also various attributes may be tagged to audio files. The present invention can generate a customized playlist for a user based upon audio file content and the attached attributes. Hence making the music listening experiences easy and customized.
Although the present invention has been described with reference to specific embodiments, these embodiments are illustrative only and not limiting. Many other applications and embodiments of the present invention will be apparent in light of this disclosure and the following claims.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4677466 | Lert et al. | Jun 1987 | A |
| 4843562 | Kenyon et al. | Jun 1989 | A |
| 5210820 | Kenyon | May 1993 | A |
| 5432852 | Leighton et al. | Jul 1995 | A |
| 5612729 | Ellis et al. | Mar 1997 | A |
| 5862260 | Rhoads | Jan 1999 | A |
| 5918223 | Blum et al. | Jun 1999 | A |
| 5960388 | Nishiguchi et al. | Sep 1999 | A |
| 5987525 | Roberts et al. | Nov 1999 | A |
| 6061680 | Scherf et al. | May 2000 | A |
| 6154773 | Roberts et al. | Nov 2000 | A |
| 6161132 | Roberts et al. | Dec 2000 | A |
| 6230192 | Roberts et al. | May 2001 | B1 |
| 6230207 | Roberts et al. | May 2001 | B1 |
| 6240459 | Roberts et al. | May 2001 | B1 |
| 6304523 | Jones et al. | Oct 2001 | B1 |
| 6330593 | Roberts et al. | Dec 2001 | B1 |
| 6453252 | Laroche | Sep 2002 | B1 |
| 6505160 | Levy et al. | Jan 2003 | B1 |
| 6512796 | Sherwood | Jan 2003 | B1 |
| 6539395 | Gjerdingen et al. | Mar 2003 | B1 |
| 6570991 | Scheirer et al. | May 2003 | B1 |
| 6571144 | Moses et al. | May 2003 | B1 |
| 6675174 | Bolle et al. | Jan 2004 | B1 |
| 6829368 | Meyer et al. | Dec 2004 | B2 |
| 20020101989 | Markandey et al. | Aug 2002 | A1 |
| 20030028796 | Roberts et al. | Feb 2003 | A1 |
| 20030046283 | Roberts | Mar 2003 | A1 |
| 20030086341 | Wells et al. | May 2003 | A1 |
| 20030135513 | Quinn et al. | Jul 2003 | A1 |
| 20040074378 | Allamanche et al. | Apr 2004 | A1 |
| 20040143349 | Roberts et al. | Jul 2004 | A1 |
| 20040172411 | Herre et al. | Sep 2004 | A1 |
| 20040267522 | Allamanche et al. | Dec 2004 | A1 |
| 20050017879 | Linzmeier et al. | Jan 2005 | A1 |
| Number | Date | Country |
|---|---|---|
| WO 9930488 | Jun 1999 | WO |
| WO0120483 | Mar 2001 | WO |
| WO0120483 | Mar 2001 | WO |
| WO0137465 | May 2001 | WO |
| WO0137465 | May 2001 | WO |
| WO02065782 | Aug 2002 | WO |
| WO02077966 | Oct 2002 | WO |
| WO02077966 | Oct 2002 | WO |
| WO02093823 | Nov 2002 | WO |
| WO03067466 | Aug 2003 | WO |
| WO03067466 | Aug 2003 | WO |
| WO2003096337 | Nov 2003 | WO |
| WO2003096337 | Nov 2003 | WO |
| WO2004044820 | May 2004 | WO |
| WO2004077430 | Sep 2004 | WO |
| WO2004081817 | Sep 2004 | WO |
