Patent Application
20120076310
This invention relates to pattern recognition and identification in media files, and more particularly to identifying repeated material in media signals, particularly audio signals, across one or more media streams.
Copyright holders, such as for music or video content, are generally entitled to compensation for each instance that their song or video is played. For music copyright holders in particular, determining when their songs are played on any of thousands of radio stations, both over the air, and now on the internet, is a daunting task. Traditionally, copyright holders have turned over collection of royalties in these circumstances to third party companies who charge entities who play music for commercial purposes a subscription fee to compensate their catalogue of copyright holders. These fees are then distributed to the copyright holders based on statistical models designed to compensate those copyright holders according which songs are receiving the most play. These statistical methods have only been very rough estimates of actual playing instances based on small sample sizes.
U.S. Pat. No. 6,990,453 issued Jan. 4, 2006 describes a system and method for comparing unknown media samples from a media stream such as a radio stations signal against a database of known media files such as songs, in order to track the instances of play for known songs. Unfortunately, much of the content of a media stream is not previously known for a variety of reasons. For example unique audio such as talk shows, the conversations or introductions of disk jockeys, or DJs, and other similar audio, represent unique audio that will not be recognizable.
There may be other unrecognized audio however, that may be of interest to a system for monitoring audio streams, and may in fact be associated with a copyright holder who should be compensated. Such non-recognized audio of interest could be a previously unindexed song, or a commercial which may use copyrighted music, or other recognized and repeated audio segments. These non-recognized audio segments may be repeated inside a single media stream or may be repeated across multiple media streams, such as a regional commercial playing on multiple radio stations.
What is needed is a system and method for recognizing repeated segments or samples in one or more otherwise non-recognized media streams, where the system and method are capable of matching samples against previously fingerprinted or index samples to find occurrences of repeated unrecognized media.
Accordingly, the present application describes a method of recognizing repeated audio material within at least one media stream without prior knowledge of the nature of the repeated material. The method including creating a screening database from the media stream or streams taking an unknown sample audio fragment from the media stream, finding matching fragments within the media stream and determining if the unknown sample matches any samples in the screening database.
In another embodiment, a system for recognizing repeated segments of non-recognized media content in at least one source of non-recognized media content, the system is described, the system including a candidate manager receiving the non-recognized media and associating an identifier with samples of the non-recognized media. The system further includes a fingerprint generator operable to create fingerprints for non-recognized media segments, and an media search engine connected to the candidate manager and the fingerprint generator, the media search engine able to compare fingerprints of non-recognized media against a database of previously stored non-recognized media fingerprints to find repeated segments within the non-recognized media content.
The foregoing has outlined rather broadly the features and technical advantages of the present invention in order that the detailed description of the invention that follows may be better understood. Additional features and advantages of the invention will be described hereinafter which form the subject of the claims of the invention. It should be appreciated by those skilled in the art that the conception and specific embodiments disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present invention. It should also be realized by those skilled in the art that such equivalent constructions do not depart from the spirit and scope of the invention as set forth in the appended claims. The novel features which are believed to be characteristic of the invention, both as to its organization and method of operation, together with further objects and advantages will be better understood from the following description when considered in connection with the accompanying figures. It is to be expressly understood, however, that each of the figures is provided for the purpose of illustration and description only and is not intended as a definition of the limits of the present invention.
For a more complete understanding of the present invention, and the advantages thereof, reference is made to the following descriptions taken in conjunction with the accompanying drawing, in which:
It is now common practice to use automated methods to identify pre-recorded material contained in audio signals such as recordings of radio or TV broadcast, or recordings of performances of material in public places such as night clubs. Regardless of the techniques used, these methods require prior access to the material to be identified, so that the signal can be matched against known content in a reference database. For much material, this is not a problem, as it may have been commercially available for some time, as in the case of music CD's. However, a significant percentage of audio signals consist of material that may not be readily available, such as music before a commercial release date, advertising material, or music written for the purposes of radio station identification, for example.
This creates two problems for those involved in accurately quantifying the content of audio signals:
Audio may be identified and segmented directly, for example using the technique by Wang (U.S. Provisional Patent Application Ser. No. 60/563,372, which is hereby entirely incorporated by reference), in which audio that is identified out of a database is segmented into known regions, with the remaining regions being left as Non-Recognized Audio (NRA). Wang's method is limited to recognizing content that is already in a database, and cannot identify and segment material not contained in a database.
To overcome this limitation, a “sifting” method is used to examine non-recognized segments of audio (NRA) from monitored media streams and to check if they match other segments or portions of segments from the recent past. Once such matches have been found, they may be segmented and included into the reference database, thus allowing databases to be created without any prior access to the target material of interest.
A system of creating a database of items of interest in NRA streams is shown in
The significant NRA segments, those having multiple matches in the monitored audio streams, may need to be sent to be positively identified and catalogued. Identifying the significant NRA segment may require that it be sent to a human operator who will listen to the audio to make a positive identification. The human operator will identify the audio and enter the necessary information to allow the significant NRA segment to be added to the database of know audio content as is described in the other applications incorporated herein by reference.
One method for recognizing repeated material is described with respect to
A screening database containing unknown (NRA) audio program material from monitored media streams in the recent past is created for screening, wherein the raw audio recordings are processed to extract fingerprint features and indexed into a searchable database. Methods for preparing such a database are disclosed in Wang and Smith (described in International Publication Number WO 02/11123 A2, entitled System and Methods for Recognizing Sound and Music Signals in High Noise and Distortion and claiming priority to U.S. Provisional Application No. 60/222,023 filed Jul. 31, 2000 and U.S. application Ser. No. 09/839,476 filed Apr. 20, 2001, (hereinafter “Wang and Smith”), the contents of which are herein incorporated by reference); or Wang and Culbert (described in International Publication No. WO 03/091990 A1, entitled “Robust and Invariant Audio Pattern Matching” and claiming priority to U.S. Provisional Application 60/376,055 filed Apr. 25, 2002, the contents of which are incorporated herein by reference (hereinafter “Wang and Culbert”)). The use of these particular methods from “Wang and Smith” or “Wang and Culbert” are illustrative and not to be construed as limiting.
To process the automatic segmentation, short probe fragments from the unknown media stream material are submitted for recognition to audio search engine 16 from
If a probe fragment F0(0) is recognized, the matching results F0,k(0) (where k is a match index) from the screening database are grouped in a matching fragment list. The task is to discover the boundaries and quality of matches of matching segments. To this end, adjacent probe fragments F0(t), where t is a time offset from F0(0), are submitted for recognition and their corresponding matching fragment lists are retrieved. The corresponding matching fragment lists are scanned for continuity, i.e., where adjacent probe fragments map onto substantially adjacent matching fragments. A chain of adjacent probe fragments may map onto one or more matching parallel chains of fragments. Such a bundle of parallel chains forms a candidate group. Each matching parallel chain is extended as far as possible in both directions in time to form a maximal matching parallel chain. A candidate segmentation point is where a plurality of matching parallel chains substantially simultaneously end or begin.
The set of maximal matching parallel chains could have different lengths and also different endpoints. This may be due to hierarchical nesting of repeated program segments. Differential segmentation could be due to different versions of a song or commercial. Alternatively, some repeated material could be embedded into repeated programs: often radio programs are aired multiple times throughout the day.
One way to determine a hierarchy of segmentation is to weight parallel chains at each fragment according to the number of matching elements in its matching fragment list. The set of parallel chains with the highest weighting is most likely to be an atomic program segment, such as a song or advertisement. Parallel chains with the next highest weightings could be due to repeated radio programs containing the atomic program segments, such as for hourly newscasts or cyclic airing of top-40 hit songs, for example. Highly-weighted parallel chains are good candidates to be atomic program segments and may be promoted to be published in a reference database for recognition of ads or songs. The criteria for publication may include such parameters as the number of candidates in the group (how many times the material has been repeated), and the exactness of the correlation between the candidates, e.g., choosing the segment with the best overall pair-wise mutual scores against other elements of its matching fragment lists. Once published, the source media stream that provided the original audio samples corresponding to the best matching exemplar of repeated material may be copied to provide a contiguous audio sample. The reason that a “best” example may need to be identified is typically because some repeated material, such as a musical track, may be overlaid with non-repeated material, such as a program presenter talking over the music. The “best” candidate will be the one with the least amount of such spurious content.
Returning to
Method 20, in process 27, then determines whether there is any old NRA that has not been matched that is ready for purging. As much of the audio stream will be unique audio that will never be matched, such as live talk shows, radio promotions, or simply the talking of radio personalities or DJs, this information needs to be purged from the system to make room for new NRA being processed. If there is NRA ready for purging, as determined from the timestamp, the availability of memory for new NRA content, or a combination of these or other factors, the method passes to process 28 which purges the old NRA. If there is no NRA for purging in process 28 or if the old NRA has been purged the process ends. It will be recognized by those skilled in the art that method 20 is a continual process that constantly attempts to recognize new NRA as the processed NRA is passed through the other processed in the method.
The above process may be interpreted as working on a fixed batch of unknown audio stream material. However, it may be enhanced to process data on an incremental basis. As new media stream content is captured, it is added to the screening database. The new material is also used to form probe fragments and scanned for repeated material as described above. Furthermore, old material may be removed from the screening database, thus preventing its unconstrained growth. One way to do this, according to “Wang and Smith”, is to continually regenerate the screening database using a moving window of unknown media stream material as new data arrives and old data is retired.
Referring to
Although the invention is not limited to any particular hardware system, an example of an embodiment of a computer system 30, which may or may not be distributed, for use in fingerprinting and landmarking media segments, such as a NRA segment is illustrated schematically in
In an alternative embodiment, certain of the functional objects are more tightly coupled together, while remaining less tightly coupled to other objects. For example, the landmarking and fingerprinting object can reside in a physically separate location from the rest of the computational objects. One example of this is a tight association of the landmarking and fingerprinting objects with the signal capturing process. In this arrangement, the landmarking and fingerprinting object can be incorporated as additional hardware or software embedded in, for example, a mobile phone, Wireless Application Protocol (WAP) browser, personal digital assistant (PDA), or other remote terminal, such as the client end of an audio search engine. In an Internet-based audio search service, such as a content identification service, the landmarking and fingerprinting object can be incorporated into the client browser application as a linked set of software instructions or a software plug-in module such as a Microsoft dynamic link library (DLL). In these embodiments, the combined signal capture, landmarking, and fingerprinting object constitutes the client end of the service. The client end sends a feature-extracted summary of the captured signal sample containing landmark and fingerprint pairs to the server end, which performs the recognition. Sending this feature-extracted summary to the server, instead of the raw captured signal, is advantageous because the amount of data is greatly reduced, often by a factor of 500 or more. Such information can be sent in real time over a low-bandwidth side channel along with or instead of, e.g., an audio stream transmitted to the server. This enables performing the invention over public communications networks, which offer relatively small-sized bandwidths to each user.
An embodiment of a method for finger printing and landmarking will now be described in detail with reference to audio samples, which can be NRA samples or segments, and NRA segments already indexed in a database such as database 14 from
Before recognition can be performed, a searchable sound database index must be constructed. As used herein, a database is any indexed collection of data, and is not limited to commercially available databases. In the database index, related elements of data are associated with one another, and individual elements can be used to retrieve associated data. The sound database index contains an index set for each file or recording in the selected collection or library of recordings, which may include speech, music, advertisements, sonar signatures, or other sounds. Each recording also has a unique identifier, sound ID. The sound database itself does not necessarily store the audio files for each recording, but the sound IDs can be used to retrieve the audio files from elsewhere. The sound database index is expected to be very large, containing indices for millions or even billions of files. New recordings are preferably added incrementally to the database index.
A block diagram of a preferred method 40 for constructing the searchable sound database index according to a first embodiment is shown in
To index the database, each recording in the collection is subjected to a landmarking and fingerprinting analysis that generates an index set for each audio file.
In step 42, each sound recording is landmarked using methods to find distinctive and reproducible locations within the sound recording. A preferred landmarking algorithm is able to mark the same timepoints within a sound recording despite the presence of noise and other linear and nonlinear distortion. Some landmarking methods are conceptually independent of the fingerprinting process described below, but can be chosen to optimize performance of the latter. Landmarking results in a list of timepoints {landmarkk} within the sound recording at which fingerprints are subsequently calculated. A good landmarking scheme marks about 5-10 landmarks per second of sound recording; of course, landmarking density depends on the amount of activity within the sound recording.
A variety of techniques are possible for computing landmarks, all of which are within the scope of the present invention. The specific technical processes used to implement the landmarking schemes of the invention are known in the art and will not be discussed in detail. A simple landmarking technique, known as Power Norm, is to calculate the instantaneous power at every possible timepoint in the recording and to select local maxima. One way of doing this is to calculate the envelope by rectifying and filtering the waveform directly.
Another way is to calculate the Hilbert transform (quadrature) of the signal and use the sum of the magnitudes squared of the Hilbert transform and the original signal.
The Power Norm method of landmarking is good at finding transients in the sound signal. The Power Norm is actually a special case of the more general Spectral Lp Norm in which p=2. The general Spectral Lp Norm is calculated at each time along the sound signal by calculating a short-time spectrum, for example via a Hanning-windowed Fast Fourier Transform (FFT). A preferred embodiment uses a sampling rate of 8000 Hz, an FFT frame size of 1024 samples, and a stride of 64 samples for each time slice. The Lp norm for each time slice is then calculated as the sum of the p̂ power of the absolute values of the spectral components, optionally followed by taking the −p̂ root. As before, the landmarks are chosen as the local maxima of the resulting values over time. An example of the Spectral Lp Norm method is shown in
When p=oo, the Lao norm is effectively the maximum norm. That is, the value of the norm is the absolute value of the largest spectral component in the spectral slice. This norm results in robust landmarks and good overall recognition performance, and is preferred for tonal music. Alternatively, “multi-slice” spectral landmarks can be calculated by taking the sum of p01 powers of absolute values of spectral components over multiple timeslices at fixed or variable offsets from each other, instead of a single slice. Finding the local maxima of this extended sum allows optimization of placement of the multi-slice fingerprints, described below.
Once the landmarks have been computed, a fingerprint is computed at each landmark timepoint in the recording in step 44 The fingerprint is generally a value or set of values that summarizes a set of features in the recording at or near the timepoint. In a currently preferred embodiment, each fingerprint is a single numerical value that is a hashed function of multiple features. Possible types of fingerprints include spectral slice fingerprints, multi-slice fingerprints, LPC coefficients, and cepstral coefficients. Of course, any type of fingerprint that characterizes the signal or features of the signal near a landmark is within the scope of the present invention. Fingerprints can be computed by any type of digital signal processing or frequency analysis of the signal.
To generate spectral slice fingerprints, a frequency analysis is performed in the neighborhood of each landmark timepoint to extract the top several spectral peaks. A simple fingerprint value is just the single frequency value of the strongest spectral peak. The use of such a simple peak results in surprisingly good recognition in the presence of noise; however, single-frequency spectral slice fingerprints tend to generate more false positives than other fingerprinting schemes because they are not unique. The number of false positives can be reduced by using fingerprints consisting of a function of the two or three strongest spectral peaks. However, there may be a higher susceptibility to noise if the second-strongest spectral peak is not sufficiently strong enough to distinguish it from its competitors in the presence of noise. That is, the calculated fingerprint value may not be sufficiently robust to be reliably reproducible. Despite this, the performance of this case is also good.
In order to take advantage of the time evolution of many sounds, a set of timeslices is determined by adding a set of time offsets to a landmark timepoint. At each resulting timeslice, a spectral slice fingerprint is calculated. The resulting set of fingerprint information is then combined to form one multitone or multi-slice fingerprint. Each multi-slice fingerprint is much more unique than the single spectral slice fingerprint, because it tracks temporal evolution, resulting in fewer false matches in the database index search described below. Experiments indicate that because of their increased uniqueness, multi-slice fingerprints computed from the single strongest spectral peak in each of two timeslices result in much faster computation (about 100 times faster) in the subsequent database index search, but with some degradation in recognition percentage in the presence of significant noise.
Alternatively, instead of using a fixed offset or offsets from a given timeslice to calculate a multi-slice fingerprint, variable offsets can be used. The variable offset to the chosen slice is the offset to the next landmark, or a landmark in a certain Offset range from the “anchor” landmark for the fingerprint. In this case, the time difference between the landmarks is also encoded into the fingerprint, along with multi-frequency information. By adding more dimensions to the fingerprints, they become more unique and have a lower chance of false match.
In addition to spectral components, other spectral features can be extracted and used as fingerprints. Linear predictive coding (LPG) analysis extracts the linearly predictable features of a signal, such as spectral peaks, as well as spectral shape. LPC is well known in the art of digital signal processing. For the present invention, LPC coefficients of waveform slices anchored at landmark positions can be used as fingerprints by hashing the quantized LPC coefficients into an index value,
Cepstrai coefficients are useful as a measure of periodicity and can be used to characterize signals that are harmonic, such as voices or many musical instruments. Cepstrai analysis is well known in the art of digital signal processing. For the present invention, a number of cepstrai coefficients are hashed together into an index and used as a fingerprint.
A block diagram conceptually illustrating the overall steps of an embodiment of a method 60 to compare NRA segments with NRA fingerprints, such as by audio search engine 16 in
In step 66, the sample fingerprints are used to retrieve sets of matching fingerprints stored in a database index 68, in which the matching fingerprints are associated with landmarks and identifiers of a set of NRA fingerprints. The set of retrieved file identifiers and landmark values are then used to generate correspondence pairs (step 70) containing sample landmarks (computed in step 64) and retrieved file landmarks at which the same fingerprints were computed. The resulting correspondence pairs are then sorted by an identifier, generating sets of correspondences between sample landmarks and file landmarks for each applicable file. Each set is scanned for alignment between the file landmarks and sample landmarks. That is, linear correspondences in the pairs of landmarks are identified, and the set is scored according to the number of pairs that are linearly related. A linear correspondence occurs when a large number of corresponding sample locations and file locations can be described with substantially the same linear equation, within an allowed tolerance. For example, if the slopes of a number of equations describing a set of correspondence pairs vary by ±5%, then the entire set of correspondences is considered to be linearly related. Of course, any suitable tolerance can be selected. The identifier of the set with the highest score, i.e., with the largest number of linearly related correspondences, is the winning NRA fingerprint identifier, which is located and returned in step 72.
As described further below, recognition can be performed with a time component proportional to the logarithm of the number of entries in the database. Recognition can be performed in essentially real time, even with a very large database. That is, a sample can be recognized as it is being obtained, with a small time lag. The method can identify a sound based on segments of 5-10 seconds and even as low 1-3 seconds. In a preferred embodiment, the landmarking and fingerprinting analysis, step 64, is carried out in real time as the samples being captured in step 62. Database queries (step 66) are carried out as sample fingerprints become available, and the correspondence results are accumulated and periodically scanned for linear correspondences. Thus all of the method steps occur simultaneously, and not in the sequential linear fashion suggested in
As described above, this method automatically identifies repeated material, with a time granularity that is dependent on the length of the audio samples originally submitted. This is in itself useful, however, with the refinements to the Audio Recognition Engine listed above, the granularity may be substantially improved. The method for increased time resolution of candidate material is the same as the above, except that the Audio Recognition Engine returns the position and length of a match within an audio sample, thus allowing the system to be free of the audio sample granularity, (reference “A Method of Characterizing the Overlap of Two Media Segments” patent application). The technique disclosed there looks at the support density of a number of matching overlapping time-aligned features extracted from the audio data. A region of “matching” overlap between two audio sample fragments has high density; conversely, non-matching regions have low density. A candidate segmentation point is chosen at a time offset within a matching media sample fragment demarcating a transition between high and low density overlap of features: This refinement yields segment endpoints within 100-200 milliseconds.
The system and method disclosed herein is typically implemented as software running on a computer system, with individual steps most efficiently implemented as independent software modules. Computer instruction code for the different objects is stored in a memory of one or more computers and executed by one or more computer processors. In one embodiment, the code objects are clustered together in a single computer system, such as an Intel-based personal computer or other workstation. In a preferred embodiment, the method is implemented by a networked cluster of central processing units (CPUs), in which different software objects are executed by different processors in order to distribute the computational load. Alternatively, each CPU can have a copy of all software objects, allowing for a homogeneous network of identically configured elements. In this latter configuration, each CPU has a subset of the database index and is responsible for searching its own subset of media files.
Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the invention as defined by the appended claims. Moreover, the scope of the present application is not intended to be limited to the particular embodiments of the process, machine, manufacture, composition of matter, means, methods and steps described in the specification. As one will readily appreciate from the disclosure, processes, machines, manufacture, compositions of matter, means, methods, or steps, presently existing or later to be developed that perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein may be utilized. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, compositions of matter, means, methods, or steps.
This application claims priority to U.S. Provisional Patent Application No. 60/651,010 filed Feb. 8, 2005, entitled AUTOMATIC IDENTIFICATION OF REPEATED MATERIAL IN AUDIO SIGNALS, the disclosure of which is hereby incorporated by reference in its entirety. This application also claims priority to U.S. patent application Ser. No. 11/350,254, filed on Feb. 8, 2006, the disclosure of which is hereby incorporated by reference in its entirety.
| Number | Date | Country | |
|---|---|---|---|
| 60651010 | Feb 2005 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11350254 | Feb 2006 | US |
| Child | 13309985 | US |
