Patent Application
20130121662
Embodiments of the invention relate generally to computing devices and systems, software, computer programs, applications, and user interfaces for identifying acoustic patterns, and more particularly, to determining equivalent portions of audio using spectral characteristics to, for example, synchronize audio and/or video captured at multiple cameras or different intervals of time.
When editing audio and video captured either by multiple cameras or in multiple takes of the same scene (e.g., with a single audio-video capture device), traditional media editing applications typically operate on the premise that audio portions captured at different cameras angles are coextensive with the captured video, and, thus, align at a common point in time. But this is often not the case. In practice, audio in multiple takes vary due slight variances in delivery, volume, word usage, utterances, etc. For example, the actors can ostensibly deliver the same lines in each take, but they might inevitably differ somewhat in timing. Sometimes they will actually say slightly different things as well, which varies the audio from take to take. Whereas, in multiple camera applications, the spatial arrangement of the cameras, as well as the environment, can also contribute to deviations in audio relative to some point in time. These deviations, which can be as small as a fraction of a second, can lead to two or more captured audio portions being out of synchronization as perceived, for example, by a human listener. Further, the efforts to edit audio and video captured in digitized form are usually exacerbated by the amounts of raw audio and video requiring editing. Specifically, editors typically expend much effort, usually manually, to search through significant amounts of content to find audio that can be synchronized for use in a final product.
One common technique for identifying similar video captured at capture devices 102a, 102b, and 102c is to implement time codes associated with each video (or otherwise use some sort of global synchronization signal) to synchronize both the video and audio portions. In particular, a user is usually required to manually adjust the different videos to bring their time codes into agreement. A time code normally describes the relative progression of a video images in terms of an hour, minute, second, and frame (e.g., HH:MM:SS:FR). But a drawback to using time codes to identify similar audio (e.g., to synchronize audio) requires the user to identify different video portions to a particular frame before synchronizing the audio portions. The effort to identify similar audio portions is further hindered due to the number of samples of audio sound that is captured relative to the number of video frames. Typically, for each frame of video (e.g., 30 frames per second), there are 1,600 samples of audio (e.g., 48,000 samples per second). As such, audio portions for capture devices 102a, 102b, and 102c are typically synchronized based on the video portions and their time codes, which can contribute to undesired sound delays and echoing effects. Another common technique for synchronizing the audio (and the video) captured at capture devices 102a, 102b, and 102c is to use a clapper to generate a distinctive sound during the capture of the audio and video. A clapper creates an audible sound—as a reference sound—to synchronize audio during the capture of the audio. The clapper sound is used for editing purposes and would otherwise be discarded during editing. The time codes and clapper sounds thus require effort to ensure their removal as they are intended for editing purposes and are distracting to an audience if time codes remain visible or clapper sounds remain audible in the final product. A drawback to using a clapper as noise 104 to synchronize audio is that the distance from noise and capture devices 102a, 102b, and 102c can cause delays that hinder synchronization of the audio relating to scene 108.
It would be desirable to provide improved computing devices and systems, software, computer programs, applications, and user interfaces that minimize one or more of the drawbacks associated with conventional techniques for identifying acoustic patterns to, for example, synchronize either audio or video, or both.
The invention and its various embodiments are more fully appreciated in connection with the following detailed description taken in conjunction with the accompanying drawings, in which:
Like reference numerals refer to corresponding parts throughout the several views of the drawings. Note that most of the reference numerals include one or two left-most digits that generally identify the figure that first introduces that reference number.
In view of the foregoing, acoustic pattern identifiers 210 and 260 of
As used herein, the term “spectral characteristic” refers generally, at least in one embodiment, to an attribute characteristic, property, quality or state, of audio (or a portion thereof) that can be used to determine whether two or more portions of audio are either equivalent or are not equivalent. A spectral characteristic can be numeric or otherwise, and can describe—in whole or in part—a portion of audio in terms of, or based on, frequency and/or power distribution (e.g., over frequencies). In one embodiment, the determination of a spectral characteristic can be either a sufficient step or an intermediary step for generating a spectral signature. In some examples, spectral characteristics can relate to a shape (or pattern) of frequency spectra (e.g., in terms of amplitude and frequency), or can be spectral coefficients. As used herein, the term “spectral signature” refers generally, at least in one embodiment, to a sample of a portion of audio that can be expressed in terms of a spectral characteristic, such as a spectral coefficient. In various embodiments, a degree of correlation for spectral signatures among different audio portions can be calculated to determine the similarity between samples of audio portions. As used herein, the term “spectral coefficient” refers generally, at least in one embodiment, to a representation of amplitude (e.g., a value indicating acoustic energy) at a specific frequency. Examples of spectral coefficients include Fourier series coefficients, layered (or low energy) coefficients, auto-correlation coefficients, linear prediction coefficients, and the like, as well as cepstral coefficients, such as linear prediction-based cepstral coefficients (“LPCC”), Fast Fourier Transform (“FFT”)-based cepstral coefficients, MEL cepstrum coefficients, and the like.
As used herein, the term “matched audio portions” refers generally, at least in one embodiment, to portions of audio having equivalent (or substantially equivalent) spectral signatures, or spectral coefficient-based measures, such as distances, with which to correlate different spectral signatures. Note that matched audio portions can include variant audio, such as is the case with audio portions captured during multiple takes of the same scene where actors might speak at different rates of speech, interject or omit different words, and the like. Regardless, acoustic pattern identifiers of various embodiments can be configured to correlate audio portions with variant audio to form matched audio portions based on the equivalency of, for example, spectral signatures. As used herein, the term “acoustic pattern” refers generally, at least in one embodiment, to the groupings (e.g., sequences) of either spectral signatures or spectral coefficient-based measures, such as distances, or both. Such groupings can indicate matched audio portions. In one embodiment, the magnitude of the spectral coefficient-based measures, such as distances, can be used to determine trough distances, which signify matching audio portions. In a specific embodiment, a relationship (e.g., a linear relationship) between spectral signatures and their distances provide for an acoustic pattern that is indicative of matching audio portions.
As used herein, the term “audio” refers generally, at least in one embodiment, to one or more sounds that are audible (e.g., perceived by humans), and can be of or relate to the transmission, storage, reproduction or reception of sound. For example, audio can be in the form of an audio waveform, an audio file, an audio signal, an audio clip, an audio track, and the like. As used herein, the term “video” refers generally, at least in one embodiment, to one or more images that are visible (e.g., perceived by humans), and can be of or relate to the transmission, storage, reproduction or reception of images. For example, video can be in the form of a video waveform, a video file, a video signal, a video clip, a video track, and the like. As used herein, the term “content” refers generally, at least in one embodiment, to information and/or material presented within a display, an interface, or the like, in relation to, for example, an audio and/or visual presentation of sounds and/or imagery. Examples of content include text, such as an electronic document (e.g., a document in Portable Document Format (“PDF”)), as well as audio, images, audio/video media, such as Flash® presentations, text, and the like. As such, a content file (or media file) can include a digital data file which is composed of images, sound and words for one camera angle. As used herein, the term “panel,” at least in one embodiment, can refer to displays, palettes, tabs, windows, screens, portions of an interface, and the like.
To illustrate the operation of spectral signature generator 322 and a spectral signature correlator 324, consider the following example. As is shown, spectral signature generator 322 can be configured to analyze audio (“Audio 1”) 302 and audio (“audio 2”) 304 to generate arrangements (e.g., vectors) (not shown) of spectral signatures 303 and 305, respectively. In particular, spectral signature generator 322 can generate a spectral signature 307 at each unit of time, such as spectral signatures SS1 at i−1, SS1 at i, SS1 at i+1 for audio 302, and can generate a spectral signature 309 at each unit of time, such as spectral signatures SS2 at j−1, SS2 at j, SS2 at j+1 for audio 304. In one embodiment, one spectral signature 307 and one spectral signature 309 can be generated at each 1/100th of a second. Spectral signature correlator 324 can be configured to calculate a correlation (or a degree of correlation) among spectral signatures 303 and 305. In one embodiment, spectral signature correlator 324 can determine a calculated correlation between a specific spectral signature 307 and a specific spectral signature 309, whereby groupings of calculated correlations can be indicative of matching audio portions between audio 302 and audio 304. In one embodiment, the calculated correlations between spectral signature 307 and spectral signature 309 can be distances.
In a specific embodiment, spectral coefficient calculator 325 can be configured to operate as a linear prediction-based cepstral coefficient (“LPCC”) calculator 340 to characterize portions of audio based on cepstral coefficients, such as linear prediction-based cepstral coefficients. In one example, spectral coefficient calculator 325 implementing linear prediction-based cepstral coefficient calculator 340 can generate linear prediction-based cepstral coefficients as follows. One or more of spectral audio matcher 321, spectral signature generator 323, and spectral coefficient calculator 325—either alone or in combination—can digitize audio from either audio file 390 or audio file 392, and, in some cases, subdivide the audio into frames over which linear prediction coefficients (“LPCs”) can be generated.
Linear prediction-based cepstral coefficient calculator 340 can convert the linear prediction coefficients into linear prediction-based cepstral coefficients. In some instances, spectral coefficient calculator 325 can implement the Levinson-Durbin algorithm, as is known, to generate the linear prediction-based cepstral coefficients. In at least one embodiment, linear prediction-based cepstral coefficient calculator 340 can calculate linear prediction-based cepstral coefficients in accordance with an inverse z-transform of the logarithm of the spectrum. In some cases, spectral signature generator 323 can generate a portion of the linear prediction-based cepstral coefficients using at least a frequency domain, with the linear prediction-based cepstral coefficients being in the time domain. In a specific embodiment, linear prediction-based cepstral coefficient calculator 340 can implement about 14 linear prediction-based cepstral coefficients, which can represent a spectral shape in a level-independent manner. In some instances, the linear prediction-based cepstral coefficients are quantized (or include a degree of quantization) in accordance with a k-means Vector Quantization algorithm—which is known—to form, for example, an 8-bit number to represent a spectral signature.
In one embodiment, spectral coefficient calculator 325 is configured to generate linear prediction-based cepstral coefficients as spectral signatures at a rate that can be in the fractions of a second, such as one generated linear prediction-based cepstral coefficient per 1/100th of a second. As such, spectral coefficient calculator 325 can generate 100 spectral signatures—as samples—for one second of audio in audio file 390 and audio file 392. Spectral coefficient calculator 325 provides these spectral signatures to spectral signature correlator 331, which, in some embodiments, can be configured to calculate correlations among the spectral signatures for audio file 390 and audio file 392 to form calculated correlations.
In this example, spectral signature correlator 331 includes a spectral signature distance engine 333, a pattern detector 335, and a pattern parametric manager 337. In a specific embodiment, spectral signature distance engine 333 is configured to determine a distance representing a correlation—or a degree of correlation—between multiple spectral signatures from, for example, audio files 390 and 392. As such, spectral signature correlator 331 can determine a distance that is indicative of whether spectral signatures associated with audio file 390 are equivalent (or substantially equivalent) to spectral signatures associated with audio file 392. As used herein, the term “distance,” at least in one embodiment, can refer to any measure that can be used to determine the degree of similarity between two or more spectral signatures.
In one embodiment, spectral signature distance engine 333 computes a distance from one spectral signature (“SS1”) to another spectral signature (“SS2”) as follows: Distance=Distance+log10(1+sqrt((SS1−SS2)̂2)), for summation over “N” coefficients. Further, SS1=[Ceps1(i)−means1(i)]/std(i), and SS2=[Ceps2(j)−means2(j)]/std(j), where Ceps1(i) is the ith coefficient for audio file 390, means1(i) is the ith mean for audio file 390, std(i) is the ith standard deviation, Ceps2(j) is the jth coefficient for audio file 392, means2(j) is the jth mean for audio file 392, std(j) is the jth standard deviation.
Pattern parametric manager 337, among other things, is configured to manage the determination of whether a specific distance is sufficient to deem two spectral signatures as being associated with the same portion of audio (or substantially so). For example, pattern parametric manager 337 can set a threshold below which a distance is sufficiently short enough for corresponding spectral signatures to be considered as being part of the same portion of audio. In some cases, this threshold is referred to as a trough distance.
Pattern detector 335 can be configured to detect whether a pattern of spectral signatures—or any other spectral characteristic—an be determined, whereby a pattern (i.e., an acoustic pattern) can be indicative of whether matching portions of audio can be identified. In one embodiment, pattern detector 335 can operate to detect patterns in which the distances for the portions of the audio are substantially coextensive to, for example, a linear relationship. That is, the distances indicating a match for the portions of the audio track each other as time linearly progresses (e.g., for each 1/100th of a second) for both audio files 390 and 392. Detected patterns can be output as matched audio portion(s).
Optionally, spectral audio matcher 321 can provide the matched audio portion (or identifiers thereof) to audio/video synchronizer 350 to synchronize audio and/or video at a synchronization point. In one embodiment, audio/video synchronizer 350 includes a cross-correlator 352 that is configured to perform a cross-correlation operation for synchronization purposes. The cross-correlation operation can implement known statistics to use cross-correlation to measure the similarity of two signals. Further, if the audio files in audio files 390 and 392 are offset from each other in relation to time, then the cross-correlation operation can figure out the offset for aligning—or synchronizing—audio files 390 and 392. In one embodiment, cross-correlator 352 correlates audio in a 1 second interval for each of audio files 390 and 392.
An acoustic pattern identifier implementing spectral signature distance engine 402 and pattern detector 404 can determine matching audio portions for audio captured during multiple takes of the same scene, according to at least one embodiment of the invention. To illustrate, consider the following in which two different people are interacting with each other. In this case, first grouping 414 can relate to audio (i.e., speech) generated by a first person, whereby second grouping 412 can relate to audio generated by a second person.
Note that second grouping 412 begins approximately after first grouping 414 finishes at S5 (of the X-axis). As such, the two persons speak in substantially a consecutive, non-overlapping fashion. Further, the timing between the two audio files is fairly synchronized as evidenced by, for example, the equivalent duration for first grouping 414, which spans 5 units of time in both audio 1 (e.g., S0 to S5) of the X-axis and audio 2 (e.g., S2 to S7) of the Y-axis. Audio files 1 and 2 can have a higher degree of synchronization should audio 2 begin at S0 of the Y-axis (not shown), which would be in synchronicity with audio 1 beginning at S0 of the X-axis.
Note, too, that the speech delivered in both audio 1 and audio 2 relating to first grouping 414 is shown to be spoken at approximately the same speed. For purposes of discussion, consider that the slope of a linear relationship coextensive with a diagonal line (not shown) defined by first grouping 414 is 45 degrees. In cases where audio 2 is delivered more slowly than audio 1, then one would expect the slope of the linear relationship to increase over 45 degrees because audio 2, which otherwise covers 5 units of time, would be extended to, for example, 6-8 units of time (not shown). The opposite can be true for cases in which audio 2 is delivered more quickly than audio 1.
Moreover, consider an instance in which the first person adds or omits utterances or speech in audio 2 relative to audio 1. While spectral signature distance engine 402 might generate some distances that are not within a trough distance (i.e., indicating mismatches in audio) for some samples (or spectral signatures) due to, for example, an omitted “uh” sound, pattern detector 404 can nevertheless operate to determine a pattern, such as first grouping 414, based on a tolerance for such variances in speech.
Note further that in some instances in which multiple people speak at the same time, such as in a situation in which two people in a coffee shop each give an exact, same order for coffee at the same time. Since their speech overlaps each other, spectral signature distance engine 402 might generate distances that may or may not be associated with higher values of distances (e.g., signifying less of a match). But pattern detector 404 can nevertheless operate to determine a pattern based on a tolerance for such variances in speech and/or spectral signatures. In view of the foregoing example, an acoustic pattern identifier can match audio portions for first grouping 414 regardless of the differences in the rate of speech or audio between audio 1 and audio 2, as well as the differences in utterances or speech (i.e., different portions of speech) in audio 2 relative to audio 1.
According to some examples, computer system 600 performs specific operations in which processor 604 executes one or more sequences of one or more instructions stored in system memory 606. Such instructions can be read into system memory 606 from another computer readable medium, such as static storage device 608 or disk drive 610. In some examples, hard-wired circuitry can be used in place of or in combination with software instructions for implementation. In the example shown, system memory 606 includes modules of executable instructions for implementing an operation system (“O/S”) 632, an application 636 (e.g., a host, client, web services-based, distributed (i.e., enterprise), application programming interface (“API”), program, procedure or others), and an audio synchronization point generation module 638.
The term “computer readable medium” refers, at least in one embodiment, to any medium that participates in providing instructions to processor 604 for execution. Such a medium can take many forms, including but not limited to, non-volatile media, volatile media, and transmission media. Non-volatile media includes, for example, optical or magnetic disks, such as disk drive 610. Volatile media includes dynamic memory, such as system memory 606. Transmission media includes coaxial cables, copper wire, and fiber optics, including wires that comprise bus 602. Transmission media can also take the form of acoustic or light waves, such as those generated during radio wave and infrared data communications.
Common forms of computer readable media includes, for example, floppy disk, flexible disk, hard disk, magnetic tape, any other magnetic medium, CD-ROM, any other optical medium, punch cards, paper tape, any other physical medium with patterns of holes, RAM, PROM, EPROM, FLASH-EPROM, any other memory chip or cartridge, carrier wave, or any other medium from which a computer can read.
In some examples, execution of the sequences of instructions can be performed by a single computer system 600. According to some examples, two or more computer systems 600 coupled by communication link 620 (e.g., LAN, PSTN, or wireless network) can perform the sequence of instructions in coordination with one another. Computer system 600 can transmit and receive messages, data, and instructions, including program code (i.e., application code) through communication link 620 and communication interface 612. Received program code can be executed by processor 604 as it is received, and/or stored in disk drive 610, or other non-volatile storage for later execution. In one embodiment, system 600 is implemented as a hand-held device But in other embodiments, system 600 can be implemented as a personal computer (i.e., a desk top computer) or any other computing device.
In at least some of the embodiments of the invention, the structures and/or functions of any of the above-described elements can be implemented in software, hardware, firmware, circuitry, or a combination thereof. Note that the structures and constituent elements described above, as well as their functionality, can be aggregated with one or more other structures or elements. Alternatively, the elements and their functionality can be subdivided into constituent sub-elements, if any. As software, the above-described described techniques can be implemented using various types of programming or formatting languages, frameworks, syntax, applications, protocols, objects, or techniques, including C, Objective C, C++, C#, Flex™, Fireworks®, Java™, Javascript™, AJAX, COBOL, Fortran, ADA, XML, HTML, DHTML, XHTML, HTTP, XMPP, and others. These can be varied and are not limited to the examples or descriptions provided.
The foregoing description, for purposes of explanation, used specific nomenclature to provide a thorough understanding of the invention. However, it will be apparent to one skilled in the art that specific details are not required in order to practice the invention. In fact, this description should not be read to limit any feature or aspect of the present invention to any embodiment; rather features and aspects of one embodiment can readily be interchanged with other embodiments.
Thus, the foregoing descriptions of specific embodiments of the invention are presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the invention to the precise forms disclosed; many alternatives, modifications, equivalents, and variations are possible in view of the above teachings. For the purpose of clarity, technical material that is known in the technical fields related to the embodiments has not been described in detail to avoid unnecessarily obscuring the description. Thus, the various embodiments can be modified within the scope and equivalents of the appended claims. Further, the embodiments were chosen and described in order to best explain the principles of the invention and its practical applications; they thereby enable others skilled in the art to best utilize the invention and various embodiments with various modifications as are suited to the particular use contemplated. Notably, not every benefit described herein need be realized by each embodiment of the present invention; rather any specific embodiment can provide one or more of the advantages discussed above. In the claims, elements and/or operations do not imply any particular order of operation, unless explicitly stated in the claims. It is intended that the following claims and their equivalents define the scope of the invention.
