WO2011146366A1 - Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization - Google Patents

Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization Download PDF

Info

Publication number
WO2011146366A1
WO2011146366A1 PCT/US2011/036601 US2011036601W WO2011146366A1 WO 2011146366 A1 WO2011146366 A1 WO 2011146366A1 US 2011036601 W US2011036601 W US 2011036601W WO 2011146366 A1 WO2011146366 A1 WO 2011146366A1
Authority
WO
WIPO (PCT)
Prior art keywords
vocal elements
lyrics
audio signal
alignment
reverse
Prior art date
Application number
PCT/US2011/036601
Other languages
French (fr)
Inventor
Ognjen Todic
Original Assignee
Shazam Entertainment Ltd.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shazam Entertainment Ltd. filed Critical Shazam Entertainment Ltd.
Priority to CN201180035459.XA priority Critical patent/CN103003875B/en
Priority to EP11724840A priority patent/EP2572354A1/en
Priority to CA2798134A priority patent/CA2798134A1/en
Priority to KR1020127032786A priority patent/KR101413327B1/en
Publication of WO2011146366A1 publication Critical patent/WO2011146366A1/en

Links

Classifications

    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/26Speech to text systems
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/04Segmentation; Word boundary detection
    • G10L15/05Word boundary detection
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10HELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
    • G10H1/00Details of electrophonic musical instruments
    • G10H1/0008Associated control or indicating means
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10HELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
    • G10H1/00Details of electrophonic musical instruments
    • G10H1/36Accompaniment arrangements
    • G10H1/361Recording/reproducing of accompaniment for use with an external source, e.g. karaoke systems
    • G10H1/368Recording/reproducing of accompaniment for use with an external source, e.g. karaoke systems displaying animated or moving pictures synchronized with the music or audio part
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00Speech recognition
    • G10L15/04Segmentation; Word boundary detection
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10HELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
    • G10H2210/00Aspects or methods of musical processing having intrinsic musical character, i.e. involving musical theory or musical parameters or relying on musical knowledge, as applied in electrophonic musical tools or instruments
    • G10H2210/031Musical analysis, i.e. isolation, extraction or identification of musical elements or musical parameters from a raw acoustic signal or from an encoded audio signal
    • G10H2210/046Musical analysis, i.e. isolation, extraction or identification of musical elements or musical parameters from a raw acoustic signal or from an encoded audio signal for differentiation between music and non-music signals, based on the identification of musical parameters, e.g. based on tempo detection
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10HELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
    • G10H2210/00Aspects or methods of musical processing having intrinsic musical character, i.e. involving musical theory or musical parameters or relying on musical knowledge, as applied in electrophonic musical tools or instruments
    • G10H2210/031Musical analysis, i.e. isolation, extraction or identification of musical elements or musical parameters from a raw acoustic signal or from an encoded audio signal
    • G10H2210/091Musical analysis, i.e. isolation, extraction or identification of musical elements or musical parameters from a raw acoustic signal or from an encoded audio signal for performance evaluation, i.e. judging, grading or scoring the musical qualities or faithfulness of a performance, e.g. with respect to pitch, tempo or other timings of a reference performance
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10HELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
    • G10H2220/00Input/output interfacing specifically adapted for electrophonic musical tools or instruments
    • G10H2220/005Non-interactive screen display of musical or status data
    • G10H2220/011Lyrics displays, e.g. for karaoke applications
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10HELECTROPHONIC MUSICAL INSTRUMENTS; INSTRUMENTS IN WHICH THE TONES ARE GENERATED BY ELECTROMECHANICAL MEANS OR ELECTRONIC GENERATORS, OR IN WHICH THE TONES ARE SYNTHESISED FROM A DATA STORE
    • G10H2250/00Aspects of algorithms or signal processing methods without intrinsic musical character, yet specifically adapted for or used in electrophonic musical processing
    • G10H2250/005Algorithms for electrophonic musical instruments or musical processing, e.g. for automatic composition or resource allocation
    • G10H2250/015Markov chains, e.g. hidden Markov models [HMM], for musical processing, e.g. musical analysis or musical composition
    • G10H2250/021Dynamic programming, e.g. Viterbi, for finding the most likely or most desirable sequence in music analysis, processing or composition

Definitions

  • Speech recognition (sometimes referred to as automatic speech recognition (ASR) or computer speech recognition) converts spoken words to text.
  • ASR automatic speech recognition
  • voice recognition is sometimes used to refer to speech, recognition where a recognition system is trained to a particular speaker to attempt to specifically identify a person speaking based on their unique vocal sound.
  • Speech recognition systems are generally based on Hidden Markov Models (HMM), which are statistical models that output a sequence of symbols or quantities.
  • HMM Hidden Markov Models
  • a speech signal can be viewed as a piecewise stationary signal or a short-time stationary signal, such that in a short-time, speech could be approximated as a stationary process. Speech could thus be thought of as a Markov model for many stochastic processes.
  • the HMMs output a sequence of ⁇ -dimensional real-valued vectors for each stationary signal.
  • the vectors include cepstral coefficients, which are obtained by taking a Fourier transform of a short time window of speech, de-correlating the transform, and taking the first (most significant) coefficients.
  • the HMM may have a statistical distribution that gives a likelihood for each observed vector.
  • Each word or each phoneme may have a different output distribution.
  • An HMM for a sequence of words or phonemes is made by concatenating individual trained HMMs for the separate words and phonemes.
  • Decoding of speech may be performed using a Viterbi decoder that determines an optimal sequence of text given the audio signal, expected grammar, and a set of HMMs that are trained on a large set of data.
  • a method of processing audio signals includes receiving an audio signal comprising vocal elements, and performing an alignment of the vocal elements with corresponding textual transcriptions of the vocal elements. The method further includes based on the alignment, determining timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements, and outputting a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
  • a forward alignment of the vocal elements processed in a forward direction with corresponding textual transcriptions of the vocal elements is performed, and a reverse alignment of the vocal elements processed in a reverse direction with corresponding reverse textual transcriptions of the vocal elements is performed.
  • the method includes determining forward timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements processed in the forward direction, and determining reverse timing boundary information associated with an elapsed amount of time for a duration of the portion of the vocal elements processed in the reverse direction.
  • the confidence metric is output based on a comparison between the forward timing information and the reverse timing information, for example.
  • the audio signal is a song comprising lyrics
  • the method further includes synchronizing the corresponding textual transcriptions of the vocal elements with the audio signal, and outputting time-annotated synchronized lyrics that indicate timing information of lines of the lyrics in relation to the audio signal.
  • a computer readable storage medium having stored therein instructions executable by a computing device to cause the computing device to perform functions.
  • the functions include receiving an audio signal comprising vocal elements, and performing an alignment of the vocal elements with corresponding textual transcriptions of the vocal elements.
  • the functions also include based on the alignment, determining timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements, and outputting a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
  • a system comprises a Hidden Markov Model (HMM) database that may include statistical modeling of phonemes in a multidimensional feature space (e.g. using Mel Frequency Cepstral Coefficients), an optional expected grammar that defines words that a speech decoder can recognize, a pronunciation dictionary database that maps words to the phonemes, and a speech decoder.
  • the speech decoder receives an audio signal and accesses the HMM, expected grammars, and a dictionary to map vocal elements in the audio signal to words.
  • the speech decoder further performs an alignment of the audio signal with corresponding textual transcriptions of the vocal elements, and determines timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements.
  • the speech decoder further determines a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
  • the speech decoder synchronizes textual transcriptions of the vocal elements with the audio signal, and outputs time-annotated synchronized lyrics that indicate timing boundary information of lines of lyrics in relation to the audio signal.
  • Figure 1 shows an illustrative embodiment of a system for performing speech recognition and synchronizing text to the recognized speech.
  • Figure 2 shows an illustrative embodiment of another system for performing speech recognition and synchronizing text to the recognized speech.
  • Figure 3 illustrates a conceptual diagram showing the reversing of the input lyrics.
  • Figure 4 is a conceptual illustration of an example of determining mismatches between forward and reverse alignments.
  • Figure 5 is a conceptual illustration of an example of determining outliers of synchronized or mapped lines using either forward and reverse alignments.
  • Figure 6 shows a flowchart of an illustrative embodiment of a method for processing audio signals.
  • Figure 7 shows a flowchart of another illustrative embodiment of a method for processing audio signals.
  • Figure 8 shows a flowchart of an illustrative embodiment of a method for processing audio signals in an iterative manner.
  • Figure 9 is a block diagram illustrating a hierarchical HMM training and model selection.
  • Figure 10 shows a flowchart of an illustrative embodiment of a method for adapting
  • Figure 11 is a block diagram illustrating an example parallel synchronization system.
  • Figure 12 is a block diagram of an example system for selecting an appropriate
  • Figure 13 is a block diagram of an example system for hybrid synchronization of audio and lyrics.
  • audio and a corresponding text may be synchronized (using speech recognition techniques in some examples), and a resulting timing metadata may be used in many different applications, such as, for example, to enable a contextual search of audio, browsing of audio, as well as display of text as audio is being played (e.g., subtitles, karaoke-like display of lyrics, etc.).
  • Example embodiments describe methods for obtaining the timing metadata, computing confidence flags for the time-synchronization metadata, and enhancing an automated synchronization process using confidence information. For example, information obtained in an automated manner may not always be accurate due to a possible mismatch between input audio and acoustic models, as well as inaccuracies in a transcript, and thus, a confidence measure that describes a quality of timing information is generated to enhance a quality of inaccurate timing metadata using automated or manual methods.
  • FIG. 1 shows an illustrative embodiment of a system 100 for performing automated synchronization using speech recognition techniques.
  • the system 100 receives an audio signal at an audio engine 102.
  • the audio signal may include a speech, a song or musical data, a TV signal, etc., and thus, may include spoken or sung words and accompanying instrumental music or background noise.
  • the audio engine 102 suppresses any instrumental music or background noise and outputs the spoken or sung words (e.g., vocals) to an automated speech recognition (ASR) decoder 104.
  • ASR automated speech recognition
  • the input audio signal is a musical song
  • the spoken or sung words may correspond to lyrics of the song, for example.
  • the audio engine 102 may suppress any instrumental music in the audio signal using techniques that leverage the fact that vocals are usually centered in a stereo signal and instrumentals are not. Music (or other non-vocal data) can also be suppressed using frequency analysis methods to identify regions that are harmonically rich. As an example, the audio engine 102 may process the audio signal using the Vocal Remover product from iZotope, Inc. The audio engine 102 may suppress non-vocal data so as to extract the vocal data or data representing spoken utterances of words, for example.
  • the system 100 also receives a lyrics text file corresponding to the lyrics of the audio signal at a filter 106.
  • the filter 106 cleans and normalizes the lyrics text. For example, the filter 106 may correct misspelling errors using lookup tables, modify vocalizations (e.g., words like 'heeeey', 'yeah', etc.) can be reduced to a smaller set (e.g. 'heeeey' and 'heeey' will be changed to 'heey'), perform grammatical changes (e.g., capitalize first letter of each line), and remove extraneous non-lyrical text (e.g., name of the artist and the song, tags potentially identifying musical segments such as chorus or verse).
  • modify vocalizations e.g., words like 'heeeey', 'yeah', etc.
  • vocalizations e.g., words like 'heeeey', 'yeah', etc
  • a grammar processor 108 receives the lyrics text from the filter 106, and creates "grammars" that indicate text that is expected to be in the vocals in the audio signal.
  • the lyrics text can be transformed into a sequence of words accompanied by "words" modeling instrumental (music-only) portions of the signal inserted at the beginning and end.
  • Optional instrumental and/or filler models can be inserted between words in the lyrics to account for voice rest and possible background accompaniment.
  • the ASR decoder 104 receives the vocals from the audio engine 102 and grammars from the grammar processor 108 and performs lyric synchronization.
  • the ASR decoder 104 will perform a forced- alignment of audio and lyrics, i.e., the expected response in the grammars will be mapped to corresponding words that are sung.
  • Accurate lyrics may be determined based on a source of the lyrics text. If the lyrics text is received from a trusted source, then accurate lyrics can be assumed, and forced-alignment can be used to map the lyrics to the audio signal.
  • force alignment grammars are defined so that there is no branching, i.e., only certain possible sequences of words can be recognized.
  • Timing information can be stored for a beginning and ending time for each line of lyrics in relation to elapsed amount of time of the song, for example, by including a timestamp or counter (not shown) in the system 100 or as a function of the ASR decoder 104.
  • the ASR decoder 104 may have access to a pronunciation dictionary database 110 that defines phonetic representations of a word (e.g., phonemes).
  • a pronunciation dictionary database 110 that defines phonetic representations of a word (e.g., phonemes).
  • the dictionary database 110 is illustrated separate from the system 100, in other examples, the dictionary database 110 may be a component of the system 100 or may be contained within components of the system 100.
  • the filter 106 may clean the lyrics text and prepare the lyrics for the grammar processor 108.
  • the grammar processor 108 will create expected response grammars from the cleaned lyrics. If the lyric source is not trusted, or if the lyrics text is not likely to fully match the words in the audio signal, the grammar processor 108 may create a stochastic grammar.
  • the grammar processor 108 may place all the lines of lyrics in parallel and allow any arbitrary sequence of lyric lines to be recognized.
  • the grammar processor 108 may insert optional and multiple words modeling instrumentals between words and at a beginning and an end of the grammar.
  • filler word models may be used to model occurrences of non-words (vocalizations, etc.).
  • grammars can be defined in a manner that allows for branching (e.g., any line of lyrics can follow any other line).
  • the audio engine 102 may analyze the suppressed audio signal by extracting feature vectors about every 10ms (e.g., using Mel Frequency Cepstral Coefficients or (MFCC)).
  • the ASR decoder 104 may then map the sequence of feature vectors to the expected response defined in the grammar.
  • the ASR decoder 104 will expand the word grammar created by the grammar processor 108 into a phonetic grammar by using the dictionary database 110 to expand words into phonemes.
  • the ASR decoder 104 may use a Hidden Markov Model (HMM) database 112 that statistically describes each phoneme in the features space (e.g,. using MFCC) to obtain an optimal sequence of words from the phonemes that matches the grammar of the audio signal and corresponding feature vector.
  • HMM database 112 is illustrated separate from the system 100, in other examples, the HMM database 112 may be a component of the system 100 or may be contained within components of the system 100.
  • HMMs are typically trained on a large set of relevant data; in the context of lyric synchronization that could be a large set of songs.
  • Estimation of model parameters can be performed using the Baum- Welch algorithm, for example.
  • Parameters of the model can be determined by re-estimation given a set of training examples corresponding to a particular model, for example.
  • the ASR decoder 104 may use an HMM from the database 112 to decode the audio signal using a Viterbi decoding algorithm that determines an optimal sequence of text given the audio signal, expected grammar, and a set of HMMs that are trained on a large set of data, for example.
  • the ASR decoder 104 uses the HMM database 112 of phonemes to map spoken words to a phonetic description, and uses the dictionary database 110 to map words to the phonetic description, for example.
  • the ASR decoder 104 will perform speech recognition or force alignment on the audio signal to create a sequence of word and phonetic transcriptions corresponding to speech in the audio signal.
  • the ASR decoder 104 When performing lyric synchronization, the ASR decoder 104 will also perform a timing analysis of the phonetic description.
  • a set of input lyrics text and corresponding phonetic transcriptions are as shown below in Table 1.
  • the phonetic transcription may be a standard dictionary transcription, such that, for example, the word “asleep” may be phonetically transcribed as "AH SH L IY P", and periods and spaces are used for clarity to indicate beginning/end of word transcriptions, to indicate pauses in the speech, or to indicate background instrumentals that may be heard between words. Note that for simplicity purposes, only a first three (out of N total) lines of the lyrics text are displayed in Table 1.
  • the audio signal may be matched to the input lyrics, so as to generate output lyrics as shown below in Table 2.
  • timing information may be output with the output lyrics, as shown in Table 2.
  • the timing information may indicate an elapsed amount of time from a beginning of a song from which the audio signal was obtained, or an elapsed amount of time from a beginning of the received audio signal to a beginning of the line of text (e.g., lyrics), and an elapsed amount of time from a beginning of the audio signal to an end of the line of lyrics.
  • the timing information may alternatively (or additionally) include an amount of time elapsed during a line, a word, or a phoneme of the lyrics.
  • a first line of the output lyrics may have a start time of 22 seconds and an end time of 24.4 seconds.
  • the start and end times are an elapsed amount of time from a beginning of the audio signal, for example.
  • a second line of output lyrics is shown in Table 2 to have a start and end time of 24.7 and 27 seconds
  • a third line of output lyrics is shown in Table 2 to have a start and end time of 27.4 and 30.2 seconds.
  • the ASR decoder 104 identifies an elapsed amount of time from a beginning of the audio signal to a time when vocals of the audio signal begin when the audio signal is played in a forward direction. Note that in the above example, timing information is specified at the line level, so the first line starts at 22 seconds and ends at 24.4 seconds. However, timing information may also be provided at a word level as well.
  • the ASR decoder 104 may determine timing information as a by-product of performing speech recognition. For example, a Viterbi decoder determines an optimal path through a matrix in which a vertical dimension represents HMM states and a horizontal dimension represents frames of speech (e.g., 10ms). When an optimal sequence of HMM states is determined, an optimal sequence of corresponding phonemes and words is available. Because each pass through the HMM state consumes a frame of speech, the timing information at the state/phoneme/word level is available as the output of the automated speech recognition.
  • a Viterbi decoder determines an optimal path through a matrix in which a vertical dimension represents HMM states and a horizontal dimension represents frames of speech (e.g., 10ms).
  • frames of speech e.g. 10ms
  • the ASR decoder 104 may include, have access to, or be operated according to a timer to determining the timing information, for example.
  • the system 100 in Figure 1 may perform time-synchronization of lyrics and audio in a batch mode (i.e., not in a real-time but instead by using a recording of the audio signal stored in the file) so as to create the timing information as shown in Table 2 above for a number of audio signals or songs.
  • Components of the system 100 in Figure 1 include engines, filters, processors, and decoders, any of which may include a computing device or a processor to execute functions of the components.
  • any of the components of the system 100 in Figure 1 may have functions embodied by computer software, which when executed by a computing device or processor cause the computing device or processor perform the functions of the components, for example.
  • the system 100 may include memory to store the computer software as well.
  • Figure 2 shows an illustrative embodiment of another system 200 for performing speech recognition and synchronizing text to the recognized speech.
  • the system 200 includes an audio engine 202 that receives an audio signal, suppresses instrumentals of the audio signal, and outputs vocals of the audio signal.
  • the audio engine 202 may output the vocals in a forward (direct) form and in a reverse form.
  • the forward form is the vocals as spoken naturally in a forward direction
  • the reverse form is the vocals reversed in a backwards or opposite direction.
  • the audio engine 202 may playback the audio signal in an opposite direction, for example.
  • the reverse form of the vocals may not be intelligible or understandable by a listener; however, the reverse form of the vocals can be used to further analyze the audio signal, for example.
  • the audio engine 202 may use the Sox software from Sound eXchange to reverse input audio signals.
  • the system also includes an ASR decoder 204 to receive the forward and reverse audio signals from the audio engine 202, and to perform speech recognition and lyric synchronization of the audio signals.
  • a filter 206 receives lyrics text that corresponds to lyrics of the audio signal, and the filter 206 cleans and normalizes the lyrics text to output the text in a direct or forward direction and in a reverse or backwards direction.
  • the forward words output from the filter 206 are the words of the lyrics written from left to right in a standard forward direction (as words as written in this disclosure).
  • the reverse words output from the filter 206 are the words of the lyrics written/read from right to left in a baclcwards direction, and thus, only the order of the words may be reversed, for example.
  • a grammar processor 208 receives the words of the lyrics in the forward and reverse direction, and outputs "grammars" corresponding to words in the forward and reverse directions.
  • the ASR decoder 204 receives the forward and reverse grammars from the grammar processor 208, as well as forward and reverse dictionary word to phoneme mappings for the forward and reverse grammars from a dictionary database 210 to map words to phonetic transcriptions, for example.
  • the ASR decoder 204 further receives statistical models of forward and reverse phonemes (e.g., small units of speech or sound that distinguish one utterance from another) from an HMM database 212. Acoustic (HMM) models for the reverse path will be trained on a training set of songs that were reversed, for example.
  • Either or both of the dictionary database 210 and the HMM database 212 may be components of the system 200, or may be contained within components of the system 200, in other examples.
  • the ASR decoder 204 may perform mapping or synchronization of the audio signal to the lyrics text in the forward direction and in the reverse direction, for example. When performing the synchronization, the ASR decoder 204 may further output timing information as described above. Example methods of the forward synchronization are described above with reference to Tables 1 and 2.
  • the ASR decoder 204 uses the reverse audio, reverse grammar, reverse phonetic dictionary (e.g., the word "asleep" is phonetically transcribed as 'P IY L SH AH' in the reverse phonetic dictionary), and reverse HMMs (e.g., each phoneme will be trained on reversed audio data, and thus, a model for phoneme 'ah' in forward and reverse HMM set would be different).
  • Table 3 illustrates reverse input lyrics and reverse phonetic transcriptions of the lyrics in Table 1.
  • the reverse input lyrics shown in Table 3 are the reverse input lyrics of Table 1.
  • a first line of the audio signal is the last line of the audio signal in Table 1.
  • the lines of the lyrics are in reverse order, and also, the words in the lines are in reverse order (e.g., reversed from the order in Table 1).
  • the corresponding phonetic transcription of lyrics, mapped via the reverse dictionary database 210 are also in reverse order (e.g., read from right to left in reverse order). Note that for simplicity only the last 3 lines of lyrics (out of N total) are displayed in the example.
  • Figure 3 illustrates a conceptual diagram showing the reversing of the input lyrics.
  • Line N in the forward direction becomes a first line in the reverse direction (Line 1 R )
  • Line N-l in the forward direction becomes a second line in the reverse direction (Line 2 R )
  • Line 1 in the forward direction becomes a the last line in the reverse direction (Line N R ), for example.
  • timing information may be output with the output lyrics in the reverse direction that may indicate an elapsed amount of time from a beginning of the received reversed audio signal.
  • the timing information may be output as an elapsed amount of time from a beginning of the audio signal to a beginning of the line of lyrics (line start time), and an elapsed amount of time from a beginning of the audio signal to an end of the line of lyrics (line end time).
  • a first line of the reverse output lyrics may have a start time of 197.8 seconds and an end time of 200.6 seconds.
  • the start and end times are an elapsed amount of time from a beginning of the reversed audio signal, for example.
  • a second line of reverse output lyrics is shown in Table 4a to have a start and end time of 202.5 and 203.3 seconds
  • a third line of reverse output lyrics is shown in Table 4a to have a start and end time of 203.6 and 206 seconds.
  • the ASR decoder 204 identifies an elapsed amount of time from a beginning of the reverse audio signal to a time when vocals of the reverse audio signal begin when the audio signal is played in a reverse direction. Note that in the above example, timing information is specified at the line level, so the line N-2 starts at 197.8 seconds and ends at 200.6 seconds. However, timing information may also be provided at a word level as well.
  • the ASR decoder 204 outputs the reverse output lyrics to a word and time reverter 214.
  • the outputs of the reverse lyrics are W N- R that indicates the reversed lines/words and T - I R that indicates the corresponding mapped timing of the lines/words.
  • the word and time reverter 214 will reverse or put the lines/words from the reverse output back to a forward direction according to Equation (1) below.
  • the output of the word and time reverter 214 is which indicates reversed output text of the reverse alignment.
  • the timing information for start of a line (or word), i can be computed as:
  • Ti RR Ttotai - T N-i R Equation (2) where T to tai is a duration of the song or audio signal and Tj R is an end time of the line in reversed synchronized lyrics.
  • a total duration of the song, T to tai is 228 seconds.
  • Table 4b below shows example data as the output of the word and time reverter 214.
  • the ASR decoder 204 may output the forward synchronized lyrics and corresponding timing information, and the "reversed" reverse synchronized lyrics and timing information to a confidence score engine 216.
  • the confidence score engine 216 computes confidence flags or scores for the timing information using a mismatch between the forward and reverse alignment.
  • the confidence score engine 216 compares a difference between the forward and reverse timing information to a predefined threshold, and marks the line as a low or high confidence line in accordance with the comparison.
  • Line timing information may be defined as T n BP where n is the line index, B defines a boundary type (S for start time, E end time) and P defines pass type (F for forward, R for reverse), then a start mismatch for line n is defined as:
  • mismatch metrics can then be compared to a predefined threshold to determine if the line should be flagged as a low or high confidence line.
  • Figure 4 is a conceptual illustration of an example of determining mismatches between the forward and reverse alignments.
  • start and end mismatch metrics would have values of zero for line boundaries of the first and last lines.
  • the start mismatch metric for the second line would have a value of zero, however, the end
  • the value of MM 2 E would be compared to a threshold value, and if 1.5 seconds exceeds the threshold value, then the second line of the lyrics would be flagged as a low confidence line. The second line of the forward and/or reversed aligned lyrics could be flagged.
  • the threshold value may be any value, for example such as about one second, and may depend to some extent on a type of the audio signal.
  • the threshold may be dynamic, such that for faster songs where lines of lyrics may be shorter in length, the threshold may be decreased.
  • the threshold for the confidence flag may be determined using techniques that minimize classification errors based on an example training set. For example, a number of false positives and or false negatives (i.e., where a line has correct boundaries but has been marked with low confidence, or has incorrect boundaries and has been marked with a high confidence) may be used as a training set.
  • a cost function may be used be when determining the threshold to minimize errors that may be more relevant for a specific application, for example, to minimize a number of bad boundaries that are flagged as good (in a case where accuracy is desired) or to minimize a number of good boundaries that are flagged as bad (in a case where minimizing additional processing cost is desired).
  • the above example uses lines of lyrics, however, the mismatch metrics may also be used at any granularity level of content, such as words or phonemes.
  • the confidence score engine 216 may also analyze forward (or reverse) recognition results and determine a probability metric of line duration given a distribution of durations of all lines in the song or audio signal. This metric leverages the symmetric notion of modern western songs and computes a probability that a duration of a specific line fits a line duration model for a song or audio signal, for example. Given the duration of each line as determined in the automated alignment process (e.g., taken from the forward and/or reverse alignment), a parametric model of line duration can be estimated by calculating a mean and standard deviation of line duration.
  • a distance from the mean duration is larger than a threshold, e.g., two standard deviations, the line is flagged as a low-confidence line.
  • a value of the threshold may differ, and may be dynamic, based on an application or desired level of accuracy of the timing boundary information, for example.
  • Table 5 below illustrates computing line duration, mean, and standard deviation using the examples above in Tables 1-2 for the forward alignment.
  • a line is marked as a low confidence line if the distance to the mean (or difference between the line duration and the mean) is greater than one standard deviation.
  • a confidence score may also be computed and output from the confidence score engine 216 on a word level, in addition to or rather than on a line level, for example.
  • the confidence score engine 216 may create a model of a line duration, and estimate a probability that the line is an outlier from the model based on a comparison of line durations.
  • An outlier may indicate that the line was incorrectly processed during speech recognition, for example.
  • the HMM models are generally not trained on the exact input audio signal, but rather are trained on training data. Thus, input audio signals may differ from those used to train the HMM models, which can result in errors during speech recognition or force-alignment.
  • methods are provided for computing confidence scores or metrics that include performing a comparison of alignment in forward and reverse directions, and performing line-duration confidence measures, for example.
  • FIG. 5 is a conceptual illustration of an example of determining outliers of synchronized or mapped lines using either the forward and reverse alignments.
  • Lines 1, 2, N-l, and N each have substantially equal timing information.
  • Line 3 has timing information T 3 (or length) that may differ by more than a threshold amount from the length of Line 1, Ti, or from the length of Line 2, T 2 .
  • T 3 timing information
  • Line 3 may be marked as an outlier using the line duration comparison.
  • estimation of line duration distribution may be constrained to lines of lyrics that belong to a same type of music segment (e.g., chorus only) as the line for which confidence is being estimated.
  • a song may be divided based on segments of the song (verse, chorus, bridge), and a value used for line duration, and thus, values of mean and standard deviation used to determine a confidence score, can be taken from a respective segment. For instance, when determining a confidence score of a line from the chorus, line durations values of lyrics corresponding to the chorus may be used.
  • the system 200 thus may output synchronized audio/lyrics in a forward and reverse direction, timing boundary information of words or lines of the lyrics in relation to the audio signal, and a confidence score/flag indicating how confident or reliable that the timing boundary information or content of the lyrics may be considered.
  • the confidence score may be determined in a number of ways, for example, based on comparison of forward and reverse timing boundary information, using line duration comparisons, using comparisons of multiple alignments performed with multiple HMMs, etc.
  • the system 200 may include or output the data to a database, and thus, the system 200 may process songs or audio signals in a batch mode to create a set of timed-annotated lyrics from a set of music and lyric files.
  • the system 200 may further use speech recognition techniques to map expected textual transcriptions of the audio signal to the audio signal.
  • correct lyrics are received and are taken as the textual transcriptions of the vocal elements in the audio signal (so that speech recognition is not needed to determine the textual transcriptions), and a forced alignment of the lyrics can be performed to the audio signal to generate timing boundary information, for example.
  • FIG. 6 shows a flowchart of an illustrative embodiment of a method 600 for processing audio signals. It should be understood that for this and other processes and methods disclosed herein, the flowchart shows functionality and operation of one possible implementation of present embodiments.
  • each block may represent a module, a segment, or a portion of program code, which includes one or more instructions executable by a processor for implementing specific logical functions or steps in the process.
  • the program code may be stored on any type of computer readable medium, for example, such as a storage device including a disk or hard drive.
  • the computer readable medium may include non-transitory computer readable medium, for example, such as computer-readable media that stores data for short periods of time like register memory, processor cache and Random Access Memory (RAM).
  • RAM Random Access Memory
  • the computer readable medium may also include non-transitory media, such as secondary or persistent long term storage, like read only memory (ROM), optical or magnetic disks, compact-disc read only memory (CD-ROM), for example.
  • the computer readable media may also be any other volatile or non-volatile storage systems, or other computer readable storage mediums.
  • each block in Figure 6 may represent circuitry that is wired to perform the specific logical functions in the process.
  • Alternative implementations are included within the scope of the example embodiments of the present disclosure in which functions may be executed out of order from that shown or discussed, including substantially concurrent or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art.
  • an input audio signal and corresponding lyrics text are received, as block 602.
  • the input audio signal may include both vocal elements and non- vocal elements, and may be a musical track or song, for example, or only a portion of a musical track or song.
  • instrumentals or non-vocals may be suppressed, as shown at block 604.
  • Timing boundary information can then be determined that is associated with an elapsed amount of time for a duration of a portion of the vocal elements, as shown at block 608.
  • a confidence metric may then be output that indicates a level of certainty for the timing boundary information for the duration of the portion of the vocal elements, as shown at block 610.
  • the confidence metric may be determined in any number of ways, for example, such as by comparing line durations of the vocal elements to search for outliers, by comparing a forward and reverse alignment output, by comparing alignments performed in parallel or serial and using different HMMs. Other examples are possible as well.
  • Figure 7 shows a flowchart of another illustrative embodiment of a method 700 for processing audio signals.
  • an input audio signal and corresponding lyrics text are received, as block 702.
  • the input audio signal may include both vocal elements and non-vocal elements, and may be a musical track or song, for example, or only a portion of a musical track or song.
  • instrumentals or non-vocals
  • forward and reverse grammars are determined from the lyrics text, as shown at block 706.
  • a forward alignment of the grammars for the lyrics text processed in a forward direction with corresponding phonetic transcriptions of the vocal elements is performed, as shown at block 708.
  • a duration of a line, word, or phoneme of the grammars corresponding to the lyrics text is determined.
  • the duration may indicate an elapsed amount of time from a beginning of the input audio signal to an end of the line of grammars, or an elapsed amount of time from a beginning of the line of grammars to the end of the line of grammars, for example.
  • a reverse alignment of the grammars for the lyrics text processed in a reverse direction with corresponding phonetic transcriptions of the vocal elements is performed, as shown at block 710.
  • a duration of a line, word, or phoneme of the reverse grammars corresponding to the reverse lyrics text is determined.
  • the forward and reverse alignment may be performed in parallel (at the same time or substantially same time) or in a serial manner, for example.
  • the forward and reverse line boundaries are then compared to compute mismatches for each line/word of the lyrics, at block 712.
  • the start and end mismatch metrics described in Equations (2)-(3) are computed and compared to a threshold value.
  • a probability metric of line duration can be computed and compared to a threshold (e.g., two standard deviations of line duration), at block 722. If the metric is within the threshold, the line of lyrics is marked as a high confidence line, at block 716. If the metric is not within the threshold, the line of lyrics is marked as a low confidence line, at block 724.
  • a threshold e.g., two standard deviations of line duration
  • audio synchronized with corresponding text, timing information, and/or confidence scores of each line of text are output, at block 726.
  • the audio synchronized with corresponding text may also include time-annotations indicating a duration of a line of the text, for example.
  • the confidence scores may indicate values of any one of the metrics described herein, or may include a high or low confidence value, for example.
  • the information output from the method 700 may be used in many different applications. Examples of such applications are described below.
  • Hidden Markov models are used for automated speech recognition, and the HMMs may be trained on a large corpus of data that aims to provide a good coverage of acoustic space, as well as generalization such that models work well on unseen speech.
  • Hidden Markov Models may be trained on a large set of training data with the goal that all variations of multiple speakers are captured. Such a type of HMM is referred to as speaker independent. Alternative HMMs can be obtained when models are trained on data that corresponds to a specific speaker, and such HMMs are referred to as speaker dependent systems. Speaker dependent systems may require that a large amount of training data for a specific speaker be collected for training purposes. However, instead of training speaker dependent models, adaptation techniques can be used. For example, using a small amount of data from the speaker, the HMM can be transformed to better fit characteristics of the speaker's voice.
  • High-quality results can be achieved when using data with known transcriptions (e.g., supervised adaptation) and with a batch of data available for adaptation (e.g., static adaptation) opposed to incremental adaptation where models are adapted as more data is available.
  • Linear transformations can be used to adapt the models, in which a set of transformations is computed using a Maximum Likelihood Linear Regression that reduces a mismatch between the adaptation data and an initial model set.
  • a Maximum a Posteriori (MAP) technique can also be used to adapt HMMs, in which prior knowledge about model parameters distribution is used.
  • the methods of Figure 6 or Figure 7 may be performed in an iterative manner.
  • the methods 600 or 700 may be performed in a first iteration, and lines (or words) of the speech or lyrics that have high-confidence scores can be selected and stored.
  • the HMMs may then be adapted using the high-confidence data of the lines (or words) of the lyrics that have high-confidence scores using supervised adaptation techniques.
  • the methods 600 or 700 may be performed in a second iteration using the retrained HMM to attempt to acquire a larger number of high-confidence scores on the lines of lyrics.
  • the HMMs may be retrained again with resulting high-confidence data, and an iterative synchronization process may continue by enhancing the HMMs via adaption using high- confidence lines output from the methods 600 or 700, for example.
  • FIG. 8 shows a flowchart of an illustrative embodiment of a method 800 for processing audio signals in an iterative manner.
  • audio and lyrics are aligned using any of the methods described herein, at block 802.
  • Time-annotated audio information is output as well as confidence scores or metric values indicating a number of high-confidence lines.
  • the audio alignment process resulted in a number of high confidence lines greater than a threshold value, at block 804 (e.g., N which may be based on amount of data needed to perform supervised adaptation, e.g., more than 1 minute of audio data)
  • the HMMs are adapted and retrained using the data from the high confidence lines, at block 806.
  • the audio and lyrics may then be realigned using the retrained HMMs, for example.
  • An output of the realignment process during the second iteration may be compared to an output of the alignment process of the first iteration, and if a number of high confidence lines in the second iteration is higher, the output of the second iteration may be stored as the time-annotated audio signal.
  • methods described herein may be used to train data- specific HMMs to be used to recognize corresponding audio signals. For example, rather than using a general HMM for a given song, selection of a most appropriate model for a given song can be made. Multiple Hidden Markov models can be trained on subsets of training data using song metadata information (e.g., genre, singer, gender, tempo, etc.) as a selection criteria.
  • Figure 9 is a block diagram illustrating a hierarchical HMM training and model selection.
  • An initial HMM training set 902 may be further adapted using genre information to generate separate models trained for a hip-hop genre 904, a pop genre 906, a rock genre 908, and a dance genre 910.
  • the genre HMMs may be further adapted to a specific tempo, such as slow hip-hop songs 912, fast hip-hop songs 914, slow dance songs 916, and fast dance songs 918. Still further, these HMMs may be adapted based on a gender of a performer, such as a slow dance song with female performer 920 and slow dance song with male performer 922. Corresponding reverse models could also be trained using the training sets with reversed audio, for example.
  • a result of a one-time training process is a database of different Hidden Markov Models each of which may include metadata specifying a specific genre, tempo, gender of the trained data, for example.
  • Figure 10 shows a flowchart of an illustrative embodiment of a method 1000 for adapting HMMs using existing synchronized-lyrics data from a specific performer.
  • An input audio signal may include information (e.g., metadata) indicating a name of the song, a name of the artist of the song, etc.
  • a system (such as system 100 or 200, for example) may search a database of synchronized lyrics to determine if there exists synchronized audio and lyrics for songs by the artist of the input audio signal, at block 1002.
  • an HMM model is retrained and adapted to the audio sample of the artist, at block 1004. If there are no synchronized lyrics for a song or audio sample by the artist of the input signal, then a standard HMM is used, at block 1006, and the audio and lyric alignment is performed at block 1008 with the appropriate HMM.
  • HMMs may be enhanced by using synchronized lyrics metadata from songs that have already been processed for a specific performer (e.g., singer). If such data already exists in the system, the data may be used to perform adaptation of the HMMs before synchronization process is performed. In this manner, a speaker independent HMM can be adapted to better model characteristics of a specific speaker.
  • an input audio sample of a particular song by The Beatles may be received along with corresponding lyrics text. If a system has performed audio-lyric synchronization of ten different songs for The Beatles, the system may first adapt a generic pop type-HMM using the previously audio-lyric synchronized data. The system may then use the adapted HMM for the audio-lyric synchronization process, for example.
  • any of the data specific HMMs may be used.
  • a parallel audio and lyric synchronization process can be performed using each of the different HMMs.
  • a best result e.g., result with a least number of low confidence lines
  • a best result among all the different outputs can be selected as a final result.
  • Figure 1 1 is a block diagram illustrating a parallel audio and lyric synchronization system 1100.
  • the system 1 100 includes a number of aligners (1, 2, . . ., N), each of which receives a copy of an input audio signal and corresponding lyrics text.
  • the aligners operate to output time-annotated synchronized audio and lyrics, and may be or include any of the components as described above in system 100 of Figure 1 or system 200 of Figure 2.
  • Each of the aligners may operate using different HMMs models (such as the different HMMs described in Figure 9), and there may a number of aligners equal to a number of different possible HMMs.
  • Outputs of the aligners will include synchronized lyrics (SLi, SL 2 , . . ., SLN), timing boundary information, and a corresponding confidence score (N owConf, N 2 L OW confj ⁇ ⁇ ⁇ > N N LowConf)-
  • the confidence score may be or include any of the metrics discussed above, and may also indicate a number of low confidence lines in the synchronized lyrics.
  • a selector 1 102 may receive the outputs of the aligners and select the output that has a best result, such as an output that has a lowest number of low confidence lines, for example.
  • a best HMM model may be selected based on criteria used to assign data to a training set, and the selected HMM model may be used to align the audio and lyrics.
  • an input audio signal may include metadata indicating a type of song, genre, tempo, performer's gender, etc., and such information may be used to select a specific HMM (as described in Figure 9) to be used during speech recognition.
  • Figure 12 is a block diagram of an example system 1200 for selecting an appropriate HMM.
  • An aligner 1202 may receive an input audio signal and lyrics text.
  • the aligner 1202 may be or include any of the components of the system 100 in Figure 1 or the system 200 in Figure 2.
  • the aligner 1202 may also receive a selected HMM from an HMM selector 1204.
  • the HMM selector 1204 may also receive the input audio signal or may receive only metadata of the input audio signal (either from the aligner 1202 or independently) and can use the metadata information to select an appropriate HMM from an HMM database 1206. For example, if the audio signal that is being processed is a slow rock song, the metadata data may indicate such information and an HMM trained on slow rock songs would be selected and provided to the aligner to be used during speech recognition. To select an appropriate HMM, a back-off technique can be used in which a most specific model is sought first, and if such a model does not exist, a less specific model will be sought, etc. If no metadata about the song is known, or if no model matches the metadata, a generic HMM would be used for the synchronization.
  • criteria can be defined to segment types of songs (e.g., genre), and HMM can be generated for specific type of song, and can subsequently be appropriately selected for using during speech recognition.
  • Figure 13 is a system 1300 for hybrid synchronization of audio and lyrics.
  • the system 1300 includes an aligner 1302, which may be or include any components of the system 100 in Figure 1 or the system 200 in Figure 2, to perform audio-lyric synchronization.
  • the aligner 1302 outputs to a user interface 1304, which may enable a user to perform manual correction of lyrics that have errors in the lyrics text or timing information, for example.
  • the system 1300 enables automated synchronization of audio and lyrics and provides for manual corrections to be made.
  • the aligner 1302 may output lines of the lyrics that have been marked with low confidence (or highlight low confidence lines) to the user interface 1304 for review or correction by a user, for example.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Reverberation, Karaoke And Other Acoustics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Electrophonic Musical Instruments (AREA)

Abstract

Methods and systems for performing audio synchronization with corresponding textual transcription and determining confidence values of the timing-synchronization are provided. Audio and a corresponding text (e.g., transcript) may be synchronized in a forward and reverse direction using speech recognition to output a time-annotated audio-lyrics synchronized data. Metrics can be computed to quantify and/or qualify a confidence of the synchronization. Based on the metrics, example embodiments describe methods for enhancing an automated synchronization process to possibly adapted Hidden Markov Models (HMMs) to the synchronized audio for use during the speech recognition. Other examples describe methods for selecting an appropriate HMM for use.

Description

Methods and Systems for Performing Synchronization of Audio with
Corresponding Textual Transcriptions and Determining Confidence Values of the Synchronization
BACKGROUND
Speech recognition (sometimes referred to as automatic speech recognition (ASR) or computer speech recognition) converts spoken words to text. The term "voice recognition" is sometimes used to refer to speech, recognition where a recognition system is trained to a particular speaker to attempt to specifically identify a person speaking based on their unique vocal sound.
Speech recognition systems are generally based on Hidden Markov Models (HMM), which are statistical models that output a sequence of symbols or quantities. A speech signal can be viewed as a piecewise stationary signal or a short-time stationary signal, such that in a short-time, speech could be approximated as a stationary process. Speech could thus be thought of as a Markov model for many stochastic processes.
The HMMs output a sequence of ^-dimensional real-valued vectors for each stationary signal. The vectors include cepstral coefficients, which are obtained by taking a Fourier transform of a short time window of speech, de-correlating the transform, and taking the first (most significant) coefficients. The HMM may have a statistical distribution that gives a likelihood for each observed vector. Each word or each phoneme may have a different output distribution. An HMM for a sequence of words or phonemes is made by concatenating individual trained HMMs for the separate words and phonemes.
Decoding of speech (e.g., when an ASR is presented with a new utterance and computes a most likely source sentence) may be performed using a Viterbi decoder that determines an optimal sequence of text given the audio signal, expected grammar, and a set of HMMs that are trained on a large set of data.
SUMMARY
In one example aspect, a method of processing audio signals is provided. The method includes receiving an audio signal comprising vocal elements, and performing an alignment of the vocal elements with corresponding textual transcriptions of the vocal elements. The method further includes based on the alignment, determining timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements, and outputting a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
In one embodiment, a forward alignment of the vocal elements processed in a forward direction with corresponding textual transcriptions of the vocal elements is performed, and a reverse alignment of the vocal elements processed in a reverse direction with corresponding reverse textual transcriptions of the vocal elements is performed. In addition, the method includes determining forward timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements processed in the forward direction, and determining reverse timing boundary information associated with an elapsed amount of time for a duration of the portion of the vocal elements processed in the reverse direction. In this embodiment, the confidence metric is output based on a comparison between the forward timing information and the reverse timing information, for example.
In another embodiment, the audio signal is a song comprising lyrics, and the method further includes synchronizing the corresponding textual transcriptions of the vocal elements with the audio signal, and outputting time-annotated synchronized lyrics that indicate timing information of lines of the lyrics in relation to the audio signal.
In another example aspect, a computer readable storage medium having stored therein instructions executable by a computing device to cause the computing device to perform functions is provided. The functions include receiving an audio signal comprising vocal elements, and performing an alignment of the vocal elements with corresponding textual transcriptions of the vocal elements. The functions also include based on the alignment, determining timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements, and outputting a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
In still another example aspect, a system is provided that comprises a Hidden Markov Model (HMM) database that may include statistical modeling of phonemes in a multidimensional feature space (e.g. using Mel Frequency Cepstral Coefficients), an optional expected grammar that defines words that a speech decoder can recognize, a pronunciation dictionary database that maps words to the phonemes, and a speech decoder. The speech decoder receives an audio signal and accesses the HMM, expected grammars, and a dictionary to map vocal elements in the audio signal to words. The speech decoder further performs an alignment of the audio signal with corresponding textual transcriptions of the vocal elements, and determines timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements. The speech decoder further determines a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
In one embodiment, the speech decoder synchronizes textual transcriptions of the vocal elements with the audio signal, and outputs time-annotated synchronized lyrics that indicate timing boundary information of lines of lyrics in relation to the audio signal.
The foregoing summary is illustrative only and is not intended to be in any way limiting. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features will become apparent by reference to the drawings and the following detailed description.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 shows an illustrative embodiment of a system for performing speech recognition and synchronizing text to the recognized speech.
Figure 2 shows an illustrative embodiment of another system for performing speech recognition and synchronizing text to the recognized speech.
Figure 3 illustrates a conceptual diagram showing the reversing of the input lyrics.
Figure 4 is a conceptual illustration of an example of determining mismatches between forward and reverse alignments.
Figure 5 is a conceptual illustration of an example of determining outliers of synchronized or mapped lines using either forward and reverse alignments.
Figure 6 shows a flowchart of an illustrative embodiment of a method for processing audio signals.
Figure 7 shows a flowchart of another illustrative embodiment of a method for processing audio signals.
Figure 8 shows a flowchart of an illustrative embodiment of a method for processing audio signals in an iterative manner.
Figure 9 is a block diagram illustrating a hierarchical HMM training and model selection. Figure 10 shows a flowchart of an illustrative embodiment of a method for adapting
HMM using existing synchronized-lyrics data from a specific performer.
Figure 11 is a block diagram illustrating an example parallel synchronization system. Figure 12 is a block diagram of an example system for selecting an appropriate
HMM.
Figure 13 is a block diagram of an example system for hybrid synchronization of audio and lyrics.
DETAILED DESCRIPTION
In the following detailed description, reference is made to the accompanying drawings, which form a part hereof. In the drawings, similar symbols typically identify similar components, unless context dictates otherwise. The illustrative embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be utilized, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein. It will be readily understood that the aspects of the present disclosure, as generally described herein, and illustrated in the Figures, can be arranged, substituted, combined, separated, and designed in a wide variety of different configurations, all of which are explicitly contemplated herein.
In example embodiments, audio and a corresponding text (e.g., transcript) may be synchronized (using speech recognition techniques in some examples), and a resulting timing metadata may be used in many different applications, such as, for example, to enable a contextual search of audio, browsing of audio, as well as display of text as audio is being played (e.g., subtitles, karaoke-like display of lyrics, etc.).
Example embodiments describe methods for obtaining the timing metadata, computing confidence flags for the time-synchronization metadata, and enhancing an automated synchronization process using confidence information. For example, information obtained in an automated manner may not always be accurate due to a possible mismatch between input audio and acoustic models, as well as inaccuracies in a transcript, and thus, a confidence measure that describes a quality of timing information is generated to enhance a quality of inaccurate timing metadata using automated or manual methods.
Figure 1 shows an illustrative embodiment of a system 100 for performing automated synchronization using speech recognition techniques. The system 100 receives an audio signal at an audio engine 102. The audio signal may include a speech, a song or musical data, a TV signal, etc., and thus, may include spoken or sung words and accompanying instrumental music or background noise. The audio engine 102 suppresses any instrumental music or background noise and outputs the spoken or sung words (e.g., vocals) to an automated speech recognition (ASR) decoder 104. When the input audio signal is a musical song, the spoken or sung words may correspond to lyrics of the song, for example.
The audio engine 102 may suppress any instrumental music in the audio signal using techniques that leverage the fact that vocals are usually centered in a stereo signal and instrumentals are not. Music (or other non-vocal data) can also be suppressed using frequency analysis methods to identify regions that are harmonically rich. As an example, the audio engine 102 may process the audio signal using the Vocal Remover product from iZotope, Inc. The audio engine 102 may suppress non-vocal data so as to extract the vocal data or data representing spoken utterances of words, for example.
The system 100 also receives a lyrics text file corresponding to the lyrics of the audio signal at a filter 106. The filter 106 cleans and normalizes the lyrics text. For example, the filter 106 may correct misspelling errors using lookup tables, modify vocalizations (e.g., words like 'heeeey', 'yeah', etc.) can be reduced to a smaller set (e.g. 'heeeey' and 'heeey' will be changed to 'heey'), perform grammatical changes (e.g., capitalize first letter of each line), and remove extraneous non-lyrical text (e.g., name of the artist and the song, tags potentially identifying musical segments such as chorus or verse).
A grammar processor 108 receives the lyrics text from the filter 106, and creates "grammars" that indicate text that is expected to be in the vocals in the audio signal. The lyrics text can be transformed into a sequence of words accompanied by "words" modeling instrumental (music-only) portions of the signal inserted at the beginning and end. Optional instrumental and/or filler models can be inserted between words in the lyrics to account for voice rest and possible background accompaniment.
The ASR decoder 104 receives the vocals from the audio engine 102 and grammars from the grammar processor 108 and performs lyric synchronization. In an example where accurate lyrics are known ahead of time, the ASR decoder 104 will perform a forced- alignment of audio and lyrics, i.e., the expected response in the grammars will be mapped to corresponding words that are sung. Accurate lyrics may be determined based on a source of the lyrics text. If the lyrics text is received from a trusted source, then accurate lyrics can be assumed, and forced-alignment can be used to map the lyrics to the audio signal. Thus, using force alignment, grammars are defined so that there is no branching, i.e., only certain possible sequences of words can be recognized. Timing information can be stored for a beginning and ending time for each line of lyrics in relation to elapsed amount of time of the song, for example, by including a timestamp or counter (not shown) in the system 100 or as a function of the ASR decoder 104.
The ASR decoder 104 may have access to a pronunciation dictionary database 110 that defines phonetic representations of a word (e.g., phonemes). Although the dictionary database 110 is illustrated separate from the system 100, in other examples, the dictionary database 110 may be a component of the system 100 or may be contained within components of the system 100. The filter 106 may clean the lyrics text and prepare the lyrics for the grammar processor 108. The grammar processor 108 will create expected response grammars from the cleaned lyrics. If the lyric source is not trusted, or if the lyrics text is not likely to fully match the words in the audio signal, the grammar processor 108 may create a stochastic grammar. To create stochastic grammar, the grammar processor 108 may place all the lines of lyrics in parallel and allow any arbitrary sequence of lyric lines to be recognized. The grammar processor 108 may insert optional and multiple words modeling instrumentals between words and at a beginning and an end of the grammar. In addition, filler word models may be used to model occurrences of non-words (vocalizations, etc.). Thus, in examples of untrusted lyric sources, grammars can be defined in a manner that allows for branching (e.g., any line of lyrics can follow any other line).
The audio engine 102 may analyze the suppressed audio signal by extracting feature vectors about every 10ms (e.g., using Mel Frequency Cepstral Coefficients or (MFCC)). The ASR decoder 104 may then map the sequence of feature vectors to the expected response defined in the grammar. The ASR decoder 104 will expand the word grammar created by the grammar processor 108 into a phonetic grammar by using the dictionary database 110 to expand words into phonemes. The ASR decoder 104 may use a Hidden Markov Model (HMM) database 112 that statistically describes each phoneme in the features space (e.g,. using MFCC) to obtain an optimal sequence of words from the phonemes that matches the grammar of the audio signal and corresponding feature vector. Although the HMM database 112 is illustrated separate from the system 100, in other examples, the HMM database 112 may be a component of the system 100 or may be contained within components of the system 100.
HMMs are typically trained on a large set of relevant data; in the context of lyric synchronization that could be a large set of songs. Estimation of model parameters can be performed using the Baum- Welch algorithm, for example. Parameters of the model can be determined by re-estimation given a set of training examples corresponding to a particular model, for example.
The ASR decoder 104 may use an HMM from the database 112 to decode the audio signal using a Viterbi decoding algorithm that determines an optimal sequence of text given the audio signal, expected grammar, and a set of HMMs that are trained on a large set of data, for example. Thus, the ASR decoder 104 uses the HMM database 112 of phonemes to map spoken words to a phonetic description, and uses the dictionary database 110 to map words to the phonetic description, for example.
The ASR decoder 104 will perform speech recognition or force alignment on the audio signal to create a sequence of word and phonetic transcriptions corresponding to speech in the audio signal.
When performing lyric synchronization, the ASR decoder 104 will also perform a timing analysis of the phonetic description. In one example, a set of input lyrics text and corresponding phonetic transcriptions are as shown below in Table 1.
Figure imgf000010_0001
Table 1
The phonetic transcription may be a standard dictionary transcription, such that, for example, the word "asleep" may be phonetically transcribed as "AH SH L IY P", and periods and spaces are used for clarity to indicate beginning/end of word transcriptions, to indicate pauses in the speech, or to indicate background instrumentals that may be heard between words. Note that for simplicity purposes, only a first three (out of N total) lines of the lyrics text are displayed in Table 1.
After performing speech recognition, the audio signal may be matched to the input lyrics, so as to generate output lyrics as shown below in Table 2.
Figure imgf000011_0001
Table 2
In addition, timing information may be output with the output lyrics, as shown in Table 2. The timing information may indicate an elapsed amount of time from a beginning of a song from which the audio signal was obtained, or an elapsed amount of time from a beginning of the received audio signal to a beginning of the line of text (e.g., lyrics), and an elapsed amount of time from a beginning of the audio signal to an end of the line of lyrics. The timing information may alternatively (or additionally) include an amount of time elapsed during a line, a word, or a phoneme of the lyrics.
As shown in Table 2, a first line of the output lyrics may have a start time of 22 seconds and an end time of 24.4 seconds. The start and end times are an elapsed amount of time from a beginning of the audio signal, for example. A second line of output lyrics is shown in Table 2 to have a start and end time of 24.7 and 27 seconds, and a third line of output lyrics is shown in Table 2 to have a start and end time of 27.4 and 30.2 seconds.
To determine the timing information, the ASR decoder 104 identifies an elapsed amount of time from a beginning of the audio signal to a time when vocals of the audio signal begin when the audio signal is played in a forward direction. Note that in the above example, timing information is specified at the line level, so the first line starts at 22 seconds and ends at 24.4 seconds. However, timing information may also be provided at a word level as well.
The ASR decoder 104 may determine timing information as a by-product of performing speech recognition. For example, a Viterbi decoder determines an optimal path through a matrix in which a vertical dimension represents HMM states and a horizontal dimension represents frames of speech (e.g., 10ms). When an optimal sequence of HMM states is determined, an optimal sequence of corresponding phonemes and words is available. Because each pass through the HMM state consumes a frame of speech, the timing information at the state/phoneme/word level is available as the output of the automated speech recognition.
Alternatively, the ASR decoder 104 may include, have access to, or be operated according to a timer to determining the timing information, for example.
The system 100 in Figure 1 may perform time-synchronization of lyrics and audio in a batch mode (i.e., not in a real-time but instead by using a recording of the audio signal stored in the file) so as to create the timing information as shown in Table 2 above for a number of audio signals or songs.
Components of the system 100 in Figure 1 include engines, filters, processors, and decoders, any of which may include a computing device or a processor to execute functions of the components. Alternatively, any of the components of the system 100 in Figure 1 may have functions embodied by computer software, which when executed by a computing device or processor cause the computing device or processor perform the functions of the components, for example. Thus, although not shown, the system 100 may include memory to store the computer software as well.
Figure 2 shows an illustrative embodiment of another system 200 for performing speech recognition and synchronizing text to the recognized speech. Many of the components of the system 200 are similar to components of the system 100, and may be embodied as computer hardware or software. For example, the system 200 includes an audio engine 202 that receives an audio signal, suppresses instrumentals of the audio signal, and outputs vocals of the audio signal. The audio engine 202 may output the vocals in a forward (direct) form and in a reverse form. The forward form is the vocals as spoken naturally in a forward direction, the reverse form is the vocals reversed in a backwards or opposite direction. To output the vocals in the reverse form, the audio engine 202 may playback the audio signal in an opposite direction, for example. The reverse form of the vocals may not be intelligible or understandable by a listener; however, the reverse form of the vocals can be used to further analyze the audio signal, for example. In one example, the audio engine 202 may use the Sox software from Sound eXchange to reverse input audio signals.
The system also includes an ASR decoder 204 to receive the forward and reverse audio signals from the audio engine 202, and to perform speech recognition and lyric synchronization of the audio signals.
A filter 206 receives lyrics text that corresponds to lyrics of the audio signal, and the filter 206 cleans and normalizes the lyrics text to output the text in a direct or forward direction and in a reverse or backwards direction. The forward words output from the filter 206 are the words of the lyrics written from left to right in a standard forward direction (as words as written in this disclosure). The reverse words output from the filter 206 are the words of the lyrics written/read from right to left in a baclcwards direction, and thus, only the order of the words may be reversed, for example.
A grammar processor 208 receives the words of the lyrics in the forward and reverse direction, and outputs "grammars" corresponding to words in the forward and reverse directions. The ASR decoder 204 receives the forward and reverse grammars from the grammar processor 208, as well as forward and reverse dictionary word to phoneme mappings for the forward and reverse grammars from a dictionary database 210 to map words to phonetic transcriptions, for example. The ASR decoder 204 further receives statistical models of forward and reverse phonemes (e.g., small units of speech or sound that distinguish one utterance from another) from an HMM database 212. Acoustic (HMM) models for the reverse path will be trained on a training set of songs that were reversed, for example. Either or both of the dictionary database 210 and the HMM database 212 may be components of the system 200, or may be contained within components of the system 200, in other examples.
The ASR decoder 204 may perform mapping or synchronization of the audio signal to the lyrics text in the forward direction and in the reverse direction, for example. When performing the synchronization, the ASR decoder 204 may further output timing information as described above. Example methods of the forward synchronization are described above with reference to Tables 1 and 2.
To perform a reverse synchronization, the ASR decoder 204 uses the reverse audio, reverse grammar, reverse phonetic dictionary (e.g., the word "asleep" is phonetically transcribed as 'P IY L SH AH' in the reverse phonetic dictionary), and reverse HMMs (e.g., each phoneme will be trained on reversed audio data, and thus, a model for phoneme 'ah' in forward and reverse HMM set would be different). Table 3 below illustrates reverse input lyrics and reverse phonetic transcriptions of the lyrics in Table 1.
Figure imgf000014_0001
Table 3
The reverse input lyrics shown in Table 3 are the reverse input lyrics of Table 1. As shown in Table 3, a first line of the audio signal is the last line of the audio signal in Table 1. Thus, the lines of the lyrics are in reverse order, and also, the words in the lines are in reverse order (e.g., reversed from the order in Table 1). Further, the corresponding phonetic transcription of lyrics, mapped via the reverse dictionary database 210, are also in reverse order (e.g., read from right to left in reverse order). Note that for simplicity only the last 3 lines of lyrics (out of N total) are displayed in the example.
Figure 3 illustrates a conceptual diagram showing the reversing of the input lyrics. As shown, for the reverse lyrics, Line N in the forward direction becomes a first line in the reverse direction (Line 1R), Line N-l in the forward direction becomes a second line in the reverse direction (Line 2R), and so forth until Line 1 in the forward direction becomes a the last line in the reverse direction (Line NR), for example.
Table 4a below indicates output lyrics with corresponding output timing information. In the same manner as described above for the forward direction, timing information may be output with the output lyrics in the reverse direction that may indicate an elapsed amount of time from a beginning of the received reversed audio signal. The timing information may be output as an elapsed amount of time from a beginning of the audio signal to a beginning of the line of lyrics (line start time), and an elapsed amount of time from a beginning of the audio signal to an end of the line of lyrics (line end time).
As shown in Table 4a, a first line of the reverse output lyrics may have a start time of 197.8 seconds and an end time of 200.6 seconds. The start and end times are an elapsed amount of time from a beginning of the reversed audio signal, for example. A second line of reverse output lyrics is shown in Table 4a to have a start and end time of 202.5 and 203.3 seconds, and a third line of reverse output lyrics is shown in Table 4a to have a start and end time of 203.6 and 206 seconds.
To determine the timing information, the ASR decoder 204 identifies an elapsed amount of time from a beginning of the reverse audio signal to a time when vocals of the reverse audio signal begin when the audio signal is played in a reverse direction. Note that in the above example, timing information is specified at the line level, so the line N-2 starts at 197.8 seconds and ends at 200.6 seconds. However, timing information may also be provided at a word level as well.
Figure imgf000016_0001
Table 4a
The ASR decoder 204 outputs the reverse output lyrics to a word and time reverter 214. The outputs of the reverse lyrics are WN- R that indicates the reversed lines/words and T -I R that indicates the corresponding mapped timing of the lines/words. The word and time reverter 214 will reverse or put the lines/words from the reverse output back to a forward direction according to Equation (1) below.
W.RR = WN_.R I= 1 :N Equation (1)
The output of the word and time reverter 214 is which indicates reversed output text of the reverse alignment.
The timing information for start of a line (or word), i, can be computed as:
TiRR = Ttotai - TN-i R Equation (2) where Ttotai is a duration of the song or audio signal and TjR is an end time of the line in reversed synchronized lyrics.
In the example described herein, a total duration of the song, Ttotai, is 228 seconds. Table 4b below shows example data as the output of the word and time reverter 214.
Figure imgf000017_0001
Table 4b
The ASR decoder 204 may output the forward synchronized lyrics and corresponding timing information, and the "reversed" reverse synchronized lyrics and timing information to a confidence score engine 216. The confidence score engine 216 computes confidence flags or scores for the timing information using a mismatch between the forward and reverse alignment.
To determine a mismatch between the forward and reverse alignment, the confidence score engine 216 compares a difference between the forward and reverse timing information to a predefined threshold, and marks the line as a low or high confidence line in accordance with the comparison. Line timing information may be defined as Tn BP where n is the line index, B defines a boundary type (S for start time, E end time) and P defines pass type (F for forward, R for reverse), then a start mismatch for line n is defined as:
MMn s = abs(Tn SF - Tn SR) Equation (3)
and an end mismatch for line n is defined as:
MMn E = abs(Tn EF - Tn ER) Equation (4)
The mismatch metrics can then be compared to a predefined threshold to determine if the line should be flagged as a low or high confidence line.
Figure 4 is a conceptual illustration of an example of determining mismatches between the forward and reverse alignments. Using the above example, start and end mismatch metrics would have values of zero for line boundaries of the first and last lines. The start mismatch metric for the second line would have a value of zero, however, the end
E EF ER. EF mismatch metric would have a value of 1.5 seconds (MM„ = abs(Tn - Tn -)} x2- = 27,
T2 ER=25.5, and MM2 E = abs(27-25.5) = 1.5). The value of MM2 E would be compared to a threshold value, and if 1.5 seconds exceeds the threshold value, then the second line of the lyrics would be flagged as a low confidence line. The second line of the forward and/or reversed aligned lyrics could be flagged.
The threshold value may be any value, for example such as about one second, and may depend to some extent on a type of the audio signal. For example, the threshold may be dynamic, such that for faster songs where lines of lyrics may be shorter in length, the threshold may be decreased. The threshold for the confidence flag may be determined using techniques that minimize classification errors based on an example training set. For example, a number of false positives and or false negatives (i.e., where a line has correct boundaries but has been marked with low confidence, or has incorrect boundaries and has been marked with a high confidence) may be used as a training set.
In addition, a cost function may used be when determining the threshold to minimize errors that may be more relevant for a specific application, for example, to minimize a number of bad boundaries that are flagged as good (in a case where accuracy is desired) or to minimize a number of good boundaries that are flagged as bad (in a case where minimizing additional processing cost is desired).
The above example uses lines of lyrics, however, the mismatch metrics may also be used at any granularity level of content, such as words or phonemes. The confidence score engine 216 may also analyze forward (or reverse) recognition results and determine a probability metric of line duration given a distribution of durations of all lines in the song or audio signal. This metric leverages the symmetric notion of modern western songs and computes a probability that a duration of a specific line fits a line duration model for a song or audio signal, for example. Given the duration of each line as determined in the automated alignment process (e.g., taken from the forward and/or reverse alignment), a parametric model of line duration can be estimated by calculating a mean and standard deviation of line duration. Then, for each line, if a distance from the mean duration is larger than a threshold, e.g., two standard deviations, the line is flagged as a low-confidence line. A value of the threshold may differ, and may be dynamic, based on an application or desired level of accuracy of the timing boundary information, for example.
Table 5 below illustrates computing line duration, mean, and standard deviation using the examples above in Tables 1-2 for the forward alignment. In the example in Table 5, a line is marked as a low confidence line if the distance to the mean (or difference between the line duration and the mean) is greater than one standard deviation.
Figure imgf000019_0001
Table 5
A confidence score may also be computed and output from the confidence score engine 216 on a word level, in addition to or rather than on a line level, for example.
In other embodiments, the confidence score engine 216 may create a model of a line duration, and estimate a probability that the line is an outlier from the model based on a comparison of line durations. An outlier may indicate that the line was incorrectly processed during speech recognition, for example. The HMM models are generally not trained on the exact input audio signal, but rather are trained on training data. Thus, input audio signals may differ from those used to train the HMM models, which can result in errors during speech recognition or force-alignment.
Thus, methods are provided for computing confidence scores or metrics that include performing a comparison of alignment in forward and reverse directions, and performing line-duration confidence measures, for example.
Figure 5 is a conceptual illustration of an example of determining outliers of synchronized or mapped lines using either the forward and reverse alignments. As shown, Lines 1, 2, N-l, and N each have substantially equal timing information. However, Line 3 has timing information T3 (or length) that may differ by more than a threshold amount from the length of Line 1, Ti, or from the length of Line 2, T2. Thus, Line 3 may be marked as an outlier using the line duration comparison.
In one example, estimation of line duration distribution may be constrained to lines of lyrics that belong to a same type of music segment (e.g., chorus only) as the line for which confidence is being estimated. For example, a song may be divided based on segments of the song (verse, chorus, bridge), and a value used for line duration, and thus, values of mean and standard deviation used to determine a confidence score, can be taken from a respective segment. For instance, when determining a confidence score of a line from the chorus, line durations values of lyrics corresponding to the chorus may be used.
The system 200 thus may output synchronized audio/lyrics in a forward and reverse direction, timing boundary information of words or lines of the lyrics in relation to the audio signal, and a confidence score/flag indicating how confident or reliable that the timing boundary information or content of the lyrics may be considered. The confidence score may be determined in a number of ways, for example, based on comparison of forward and reverse timing boundary information, using line duration comparisons, using comparisons of multiple alignments performed with multiple HMMs, etc. The system 200 may include or output the data to a database, and thus, the system 200 may process songs or audio signals in a batch mode to create a set of timed-annotated lyrics from a set of music and lyric files.
The system 200 may further use speech recognition techniques to map expected textual transcriptions of the audio signal to the audio signal. Alternatively, correct lyrics are received and are taken as the textual transcriptions of the vocal elements in the audio signal (so that speech recognition is not needed to determine the textual transcriptions), and a forced alignment of the lyrics can be performed to the audio signal to generate timing boundary information, for example.
Figure 6 shows a flowchart of an illustrative embodiment of a method 600 for processing audio signals. It should be understood that for this and other processes and methods disclosed herein, the flowchart shows functionality and operation of one possible implementation of present embodiments. In this regard, each block may represent a module, a segment, or a portion of program code, which includes one or more instructions executable by a processor for implementing specific logical functions or steps in the process. The program code may be stored on any type of computer readable medium, for example, such as a storage device including a disk or hard drive. The computer readable medium may include non-transitory computer readable medium, for example, such as computer-readable media that stores data for short periods of time like register memory, processor cache and Random Access Memory (RAM). The computer readable medium may also include non-transitory media, such as secondary or persistent long term storage, like read only memory (ROM), optical or magnetic disks, compact-disc read only memory (CD-ROM), for example. The computer readable media may also be any other volatile or non-volatile storage systems, or other computer readable storage mediums. In addition, each block in Figure 6 may represent circuitry that is wired to perform the specific logical functions in the process. Alternative implementations are included within the scope of the example embodiments of the present disclosure in which functions may be executed out of order from that shown or discussed, including substantially concurrent or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art.
Initially in the method 600, an input audio signal and corresponding lyrics text are received, as block 602. The input audio signal may include both vocal elements and non- vocal elements, and may be a musical track or song, for example, or only a portion of a musical track or song. Following, as an optional step, instrumentals (or non-vocals) may be suppressed, as shown at block 604.
Then, an alignment of the vocal elements with the corresponding textual transcriptions of the vocal elements is performed, as shown at block 606. Timing boundary information can then be determined that is associated with an elapsed amount of time for a duration of a portion of the vocal elements, as shown at block 608.
A confidence metric may then be output that indicates a level of certainty for the timing boundary information for the duration of the portion of the vocal elements, as shown at block 610. The confidence metric may be determined in any number of ways, for example, such as by comparing line durations of the vocal elements to search for outliers, by comparing a forward and reverse alignment output, by comparing alignments performed in parallel or serial and using different HMMs. Other examples are possible as well.
Figure 7 shows a flowchart of another illustrative embodiment of a method 700 for processing audio signals. Initially in the method 700, an input audio signal and corresponding lyrics text are received, as block 702. The input audio signal may include both vocal elements and non-vocal elements, and may be a musical track or song, for example, or only a portion of a musical track or song. Following, as an optional step, instrumentals (or non-vocals) may be suppressed, as shown at block 704. Then forward and reverse grammars are determined from the lyrics text, as shown at block 706.
Next, a forward alignment of the grammars for the lyrics text processed in a forward direction with corresponding phonetic transcriptions of the vocal elements is performed, as shown at block 708. As part of the forward alignment, at the same time, or subsequently, a duration of a line, word, or phoneme of the grammars corresponding to the lyrics text is determined. The duration may indicate an elapsed amount of time from a beginning of the input audio signal to an end of the line of grammars, or an elapsed amount of time from a beginning of the line of grammars to the end of the line of grammars, for example.
In addition, a reverse alignment of the grammars for the lyrics text processed in a reverse direction with corresponding phonetic transcriptions of the vocal elements is performed, as shown at block 710. As part of the reverse alignment, at the same time, or subsequently, a duration of a line, word, or phoneme of the reverse grammars corresponding to the reverse lyrics text is determined. The forward and reverse alignment may be performed in parallel (at the same time or substantially same time) or in a serial manner, for example.
The forward and reverse line boundaries are then compared to compute mismatches for each line/word of the lyrics, at block 712. As one example, the start and end mismatch metrics described in Equations (2)-(3) are computed and compared to a threshold value.
Based on the comparison performed, a determination is made whether the metric is within a given threshold, at block 714. If the metric is within the threshold, the line of lyrics is marked as a high confidence line, at block 716. A high confidence line has a high reliability, certainty, or probability that the start and end time of the line highly or reliably corresponds to the vocal elements in the input audio signal. If the metric is not within the threshold, the line of lyrics is marked as a low confidence line, at block 718. A low confidence line has a low reliability, certainty, or probability that the line of grammars reliably corresponds to the vocal elements in the input audio signal.
As another example, at block 720, a probability metric of line duration can be computed and compared to a threshold (e.g., two standard deviations of line duration), at block 722. If the metric is within the threshold, the line of lyrics is marked as a high confidence line, at block 716. If the metric is not within the threshold, the line of lyrics is marked as a low confidence line, at block 724.
Following, audio synchronized with corresponding text, timing information, and/or confidence scores of each line of text are output, at block 726. The audio synchronized with corresponding text may also include time-annotations indicating a duration of a line of the text, for example. The confidence scores may indicate values of any one of the metrics described herein, or may include a high or low confidence value, for example.
The information output from the method 700 may be used in many different applications. Examples of such applications are described below.
In one example, in the system 100 of Figure 1 or the system 200 of Figure 2, Hidden Markov models are used for automated speech recognition, and the HMMs may be trained on a large corpus of data that aims to provide a good coverage of acoustic space, as well as generalization such that models work well on unseen speech.
Hidden Markov Models may be trained on a large set of training data with the goal that all variations of multiple speakers are captured. Such a type of HMM is referred to as speaker independent. Alternative HMMs can be obtained when models are trained on data that corresponds to a specific speaker, and such HMMs are referred to as speaker dependent systems. Speaker dependent systems may require that a large amount of training data for a specific speaker be collected for training purposes. However, instead of training speaker dependent models, adaptation techniques can be used. For example, using a small amount of data from the speaker, the HMM can be transformed to better fit characteristics of the speaker's voice. High-quality results can be achieved when using data with known transcriptions (e.g., supervised adaptation) and with a batch of data available for adaptation (e.g., static adaptation) opposed to incremental adaptation where models are adapted as more data is available. Linear transformations can be used to adapt the models, in which a set of transformations is computed using a Maximum Likelihood Linear Regression that reduces a mismatch between the adaptation data and an initial model set. Alternatively, a Maximum a Posteriori (MAP) technique can also be used to adapt HMMs, in which prior knowledge about model parameters distribution is used.
In an example embodiment, the methods of Figure 6 or Figure 7 may be performed in an iterative manner. The methods 600 or 700 may be performed in a first iteration, and lines (or words) of the speech or lyrics that have high-confidence scores can be selected and stored. The HMMs may then be adapted using the high-confidence data of the lines (or words) of the lyrics that have high-confidence scores using supervised adaptation techniques. For example, the methods 600 or 700 may be performed in a second iteration using the retrained HMM to attempt to acquire a larger number of high-confidence scores on the lines of lyrics. The HMMs may be retrained again with resulting high-confidence data, and an iterative synchronization process may continue by enhancing the HMMs via adaption using high- confidence lines output from the methods 600 or 700, for example.
Figure 8 shows a flowchart of an illustrative embodiment of a method 800 for processing audio signals in an iterative manner. Initially, audio and lyrics are aligned using any of the methods described herein, at block 802. Time-annotated audio information is output as well as confidence scores or metric values indicating a number of high-confidence lines. Next, if the audio alignment process resulted in a number of high confidence lines greater than a threshold value, at block 804 (e.g., N which may be based on amount of data needed to perform supervised adaptation, e.g., more than 1 minute of audio data), then the HMMs are adapted and retrained using the data from the high confidence lines, at block 806. The audio and lyrics may then be realigned using the retrained HMMs, for example.
An output of the realignment process during the second iteration may be compared to an output of the alignment process of the first iteration, and if a number of high confidence lines in the second iteration is higher, the output of the second iteration may be stored as the time-annotated audio signal.
In another example embodiment, methods described herein may be used to train data- specific HMMs to be used to recognize corresponding audio signals. For example, rather than using a general HMM for a given song, selection of a most appropriate model for a given song can be made. Multiple Hidden Markov models can be trained on subsets of training data using song metadata information (e.g., genre, singer, gender, tempo, etc.) as a selection criteria. Figure 9 is a block diagram illustrating a hierarchical HMM training and model selection. An initial HMM training set 902 may be further adapted using genre information to generate separate models trained for a hip-hop genre 904, a pop genre 906, a rock genre 908, and a dance genre 910. The genre HMMs may be further adapted to a specific tempo, such as slow hip-hop songs 912, fast hip-hop songs 914, slow dance songs 916, and fast dance songs 918. Still further, these HMMs may be adapted based on a gender of a performer, such as a slow dance song with female performer 920 and slow dance song with male performer 922. Corresponding reverse models could also be trained using the training sets with reversed audio, for example.
A result of a one-time training process is a database of different Hidden Markov Models each of which may include metadata specifying a specific genre, tempo, gender of the trained data, for example. Still further, in another example, Figure 10 shows a flowchart of an illustrative embodiment of a method 1000 for adapting HMMs using existing synchronized-lyrics data from a specific performer. An input audio signal may include information (e.g., metadata) indicating a name of the song, a name of the artist of the song, etc. A system (such as system 100 or 200, for example) may search a database of synchronized lyrics to determine if there exists synchronized audio and lyrics for songs by the artist of the input audio signal, at block 1002. If there exists synchronized lyrics for a song or audio sample by the artist of the input signal, then an HMM model is retrained and adapted to the audio sample of the artist, at block 1004. If there are no synchronized lyrics for a song or audio sample by the artist of the input signal, then a standard HMM is used, at block 1006, and the audio and lyric alignment is performed at block 1008 with the appropriate HMM. Using the method of 1000, HMMs may be enhanced by using synchronized lyrics metadata from songs that have already been processed for a specific performer (e.g., singer). If such data already exists in the system, the data may be used to perform adaptation of the HMMs before synchronization process is performed. In this manner, a speaker independent HMM can be adapted to better model characteristics of a specific speaker.
In a specific example of an application of methods in Figures 8-10, an input audio sample of a particular song by The Beatles may be received along with corresponding lyrics text. If a system has performed audio-lyric synchronization of ten different songs for The Beatles, the system may first adapt a generic pop type-HMM using the previously audio-lyric synchronized data. The system may then use the adapted HMM for the audio-lyric synchronization process, for example.
In one embodiment, during any of the methods described herein, any of the data specific HMMs (e.g., as shown in Figure 9 or enhanced as described in Figure 10) may be used. In one example, a parallel audio and lyric synchronization process can be performed using each of the different HMMs. Using the resulting confidence information, a best result (e.g., result with a least number of low confidence lines) among all the different outputs can be selected as a final result.
Figure 1 1 is a block diagram illustrating a parallel audio and lyric synchronization system 1100. The system 1 100 includes a number of aligners (1, 2, . . ., N), each of which receives a copy of an input audio signal and corresponding lyrics text. The aligners operate to output time-annotated synchronized audio and lyrics, and may be or include any of the components as described above in system 100 of Figure 1 or system 200 of Figure 2. Each of the aligners may operate using different HMMs models (such as the different HMMs described in Figure 9), and there may a number of aligners equal to a number of different possible HMMs.
Outputs of the aligners will include synchronized lyrics (SLi, SL2, . . ., SLN), timing boundary information, and a corresponding confidence score (N owConf, N2LOWconfj · · ·> NNLowConf)- The confidence score may be or include any of the metrics discussed above, and may also indicate a number of low confidence lines in the synchronized lyrics. A selector 1 102 may receive the outputs of the aligners and select the output that has a best result, such as an output that has a lowest number of low confidence lines, for example.
In another example, a best HMM model may be selected based on criteria used to assign data to a training set, and the selected HMM model may be used to align the audio and lyrics. For example, an input audio signal may include metadata indicating a type of song, genre, tempo, performer's gender, etc., and such information may be used to select a specific HMM (as described in Figure 9) to be used during speech recognition. Figure 12 is a block diagram of an example system 1200 for selecting an appropriate HMM. An aligner 1202 may receive an input audio signal and lyrics text. The aligner 1202 may be or include any of the components of the system 100 in Figure 1 or the system 200 in Figure 2. The aligner 1202 may also receive a selected HMM from an HMM selector 1204. The HMM selector 1204 may also receive the input audio signal or may receive only metadata of the input audio signal (either from the aligner 1202 or independently) and can use the metadata information to select an appropriate HMM from an HMM database 1206. For example, if the audio signal that is being processed is a slow rock song, the metadata data may indicate such information and an HMM trained on slow rock songs would be selected and provided to the aligner to be used during speech recognition. To select an appropriate HMM, a back-off technique can be used in which a most specific model is sought first, and if such a model does not exist, a less specific model will be sought, etc. If no metadata about the song is known, or if no model matches the metadata, a generic HMM would be used for the synchronization.
Thus, using the examples shown in Figures 8-12, criteria can be defined to segment types of songs (e.g., genre), and HMM can be generated for specific type of song, and can subsequently be appropriately selected for using during speech recognition.
Figure 13 is a system 1300 for hybrid synchronization of audio and lyrics. The system 1300 includes an aligner 1302, which may be or include any components of the system 100 in Figure 1 or the system 200 in Figure 2, to perform audio-lyric synchronization. The aligner 1302 outputs to a user interface 1304, which may enable a user to perform manual correction of lyrics that have errors in the lyrics text or timing information, for example. Thus, the system 1300 enables automated synchronization of audio and lyrics and provides for manual corrections to be made. In one embodiment, the aligner 1302 may output lines of the lyrics that have been marked with low confidence (or highlight low confidence lines) to the user interface 1304 for review or correction by a user, for example.
While various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for purposes of illustration and are not intended to be limiting, with the true scope and spirit being indicated by the following claims. Many modifications and variations can be made without departing from its spirit and scope, as will be apparent to those skilled in the art. Functionally equivalent methods and apparatuses within the scope of the disclosure, in addition to those enumerated herein, will be apparent to those skilled in the art from the foregoing descriptions. Such modifications and variations are intended to fall within the scope of the appended claims.

Claims

CLAIMS What is claimed is:
1. A method of processing audio signals, comprising:
receiving an audio signal comprising vocal elements;
a processor performing an alignment of the vocal elements with corresponding textual transcriptions of the vocal elements;
based on the alignment, determining timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements; and
outputting a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
2. The method of claim 1, wherein performing the alignment of the vocal elements with corresponding textual transcriptions of the vocal elements comprises performing a forward alignment of the vocal elements processed in a forward direction with corresponding textual transcriptions of the vocal elements, wherein determining timing boundary information comprises determining forward timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements processed in the forward direction, and wherein the method further comprises:
performing a reverse alignment of the vocal elements processed in a reverse direction with corresponding reverse textual transcriptions of the vocal elements;
determining reverse timing boundary information associated with an elapsed amount of time for a duration of the portion of the vocal elements processed in the reverse direction; and
based on a comparison between the forward timing information and the reverse timing information, outputting the confidence metric indicating a level of certainty for the forward timing boundary information.
3. The method of claim 2, further comprising:
determining a difference between the forward timing boundary information and the reverse timing boundary information;
making a comparison of the difference to a predefined threshold; and
based on the comparison, marking the portion of the vocal elements with a confidence level.
4. The method of claim 2, wherein the audio signal is a musical song, and wherein the portion of the vocal elements is a line of the musical song, and wherein the forward timing information and the reverse timing information each indicate a start and end time of the line of the musical song, and wherein the method further comprises comparing the forward timing information and the reverse timing information by:
comparing the start time of the forward timing information with the start time of the reverse timing information to provide a start mismatch metric; and
comparing the end time of the forward timing information and the end time of the reverse timing information to provide an end mismatch metric.
5. The method of claim 1, further comprising:
for each of a plurality of portions of the vocal elements, determining timing boundary information;
computing a statistical model for a given duration of a portion of the vocal elements based on the plurality of portions of the vocal elements; for each of a plurality of portions of the vocal elements, determining a probability that the duration fits the statistical model and comparing the probability to a threshold; and
for portions of the vocal elements that that have a probability lower than the threshold, marking the portion of the vocal elements with a low-confidence flag.
6. The method of claim 1, further comprising:
for each of a plurality of portions of the vocal elements, determining timing boundary information;
comparing the timing boundary information of each of the plurality of portions of the vocal elements amongst each other; and
based on the comparison, identifying outliers of the plurality of portions of the vocal elements.
7. The method of claim 1, wherein the audio signal comprises vocal elements and non-vocal elements, and the method further comprises suppressing the non-vocal elements.
8. The method of claim 1, wherein the audio signal is a song comprising lyrics, and wherein the method further comprises:
synchronizing the corresponding textual transcriptions of the vocal elements with the audio signal; and
outputting time-annotated synchronized lyrics that indicate timing information of lines of the lyrics in relation to the audio signal.
9. The method of claim 1, wherein performing the alignment comprises performing speech recognition on the vocal elements using a Viterbi decoder and Hidden Markov Models (HMM), and wherein the audio signal is a musical track by an artist, and the method further comprises:
accessing a database for synchronized lyrics of the artist;
adapting the HMM using the synchronized lyrics of the artist as adaptation data to produce an updated HMM; and
repeating the alignment using the updated HMM.
10. The method of claim 1, wherein the audio signal is a musical track, and wherein the portion of the vocal elements is selected from the group consisting of a line of lyrics of the musical tract and a word of lyrics of the musical track.
11. The method of claim 10, wherein outputting the confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements comprises marking the line of lyrics of the musical tract as a high or low confidence line.
12. The method of claim 1, further comprising:
performing speech recognition on the vocal elements to create a sequence of phonetic transcriptions; and
performing an alignment of the vocal elements with the phonetic transcriptions.
13. The method of claim 12, wherein performing the alignment comprises:
receiving lyrics text corresponding to the audio signal; determining grammars for the lyrics text; and
mapping the phonetic description of the vocal elements to the grammars for the lyrics text.
14. The method of claim 1, wherein performing the alignment comprises performing speech recognition on a plurality of portions of the vocal elements using a Hidden Markov Model (HMM), and wherein the method further comprises:
making a determination whether the confidence metric exceeds a predetermined threshold; and
adapting the HMM using data comprising portions of the vocal elements that have a confidence metric that does not exceed the predetermined threshold to produce an updated HMM; and
repeating the alignment using the updated HMM.
15. The method of claim 14, further comprising repeating the steps of performing the alignment, outputting the confidence metric, adapting the HMM, and repeating the alignment in an iterative manner until there is no further decrease in a number of low- confidence lines.
16. The method of claim 1, wherein performing the alignment comprises performing speech recognition on the vocal elements using a Hidden Markov Model (HMM), and the method further comprises:
training a database of HMMs on training data based on metadata information of the audio signal; and
selecting an HMM to perform the alignment based on metadata information of the audio signal.
17. The method of claim 16, wherein the metadata information indicates information selected from the group consisting of a genre, an artist, a gender, and a tempo.
18. The method of claim 1, further comprising
performing the alignment a plurality of times using different Hidden Markov Model (HMM) for each alignment;
determining the timing boundary information for each respective alignment;
determining the confidence metric for each respective alignment;
selecting an alignment that has a confidence metric indicating a highest level of certainty for the timing boundary information; and v
outputting time-annotated synchronized lyrics that indicate timing boundary information corresponding to the selected alignment, wherein the timing boundary information pertains to lines of lyrics in relation to the audio signal.
19. A computer readable storage medium having stored therein instructions executable by a computing device to cause the computing device to perform functions of: receiving an audio signal comprising vocal elements;
performing an alignment of the vocal elements with corresponding textual transcriptions of the vocal elements;
based on the alignment, determining timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements; and
outputting a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
20. The computer readable storage medium of claim 19, wherein the function of performing the alignment of the vocal elements with corresponding textual transcriptions of the vocal elements comprises performing a forward alignment of the vocal elements processed in a forward direction with corresponding textual transcriptions of the vocal elements, wherein the function of determining timing boundary information comprises determining forward timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements processed in the forward direction, and wherein the instructions are further executable to perform functions of:
performing a reverse alignment of the vocal elements processed in a reverse direction with corresponding reverse textual transcriptions of the vocal elements;
determining reverse timing boundary information associated with an elapsed amount of time for a duration of the portion of the vocal elements processed in the reverse direction; and
based on a comparison between the forward timing information and the reverse timing information, outputting the confidence metric indicating a level of certainly for the forward timing boundary information.
21. The computer readable storage medium of claim 20, wherein the instructions are further executable to perform functions of:
determining a difference between the forward timing boundary information and the reverse timing boundary information;
making a comparison of the difference to a predefined threshold; and
based on the comparison, marking the portion of the vocal elements with a confidence level.
22. The computer readable storage medium of claim 20, wherein the audio signal is a musical song, and wherein the portion of the vocal elements is a line of the musical song, and wherein the forward timing information and the reverse timing information each indicate a start and end time of the line of the musical song, and wherein the instructions are further executable to perform functions of comparing the forward timing information and the reverse timing information by:
comparing the start time of the forward timing information with the start time of the reverse timing information to provide a start mismatch metric; and
comparing the end time of the forward timing information and the end time of the reverse timing information to provide an end mismatch metric.
23. The computer readable storage medium of claim 19, wherein the instructions are further executable to perform functions of:
for each of a plurality of portions of the vocal elements, determining timing boundary information;
computing a mean value of the timing boundary information for the plurality of portions of the vocal elements;
for each of a plurality of portions of the vocal elements, determining whether the duration of the portion of the vocal elements differs from the mean value by more than a threshold; and
for portions of the vocal elements that differ from the mean value by more than a threshold, marking the portion of the vocal elements with a low-confidence probability.
24. The computer readable storage medium of claim 19, wherein the audio signal is a song comprising lyrics, and wherein the instructions are further executable to perform functions of:
synchronizing the corresponding textual transcriptions of the vocal elements with the audio signal; and
outputting time-annotated synchronized lyrics that indicate timing information of lines of the lyrics in relation to the audio signal.
25. The computer readable storage medium of claim 19, wherein the function of performing the alignment comprises performing speech recognition on the vocal elements using a Hidden Markov Model (HMM), and wherein the instructions are further executable to perform functions of:
selecting an HMM based on metadata information of the audio signal.
26. A system comprising:
a Hidden Markov Model (HMM) database that includes phonetic modeling of words; a pronunciation dictionary database that includes grammars representing words; and a speech decoder that receives an audio signal and accesses the HMM to map vocal elements in the audio signal to phonetic descriptions and accesses the pronunciation dictionary database to map the phonetic descriptions to grammars, the speech decoder further performing an alignment of the grammars with corresponding textual transcriptions of the vocal elements,
wherein the speech decoder determines timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements, and the speech decoder determines a confidence metric indicating a level of certainty for the timing boundary information for the duration of the portion of the vocal elements.
27. The system of claim 26, wherein the speech decoder receives the audio signal in a forward direction and a reverse direction and accesses the HMM to map vocal elements in the audio signal to phonetic descriptions in both the forward direction and the reverse direction and accesses the pronunciation dictionary database to map the phonetic descriptions to grammars in both the forward and the reverse direction, the speech decoder further performing the alignment of the grammars with corresponding textual transcriptions of the vocal elements in both the forward direction and the reverse direction, and
wherein the speech decoder determines forward timing boundary information associated with an elapsed amount of time for a duration of a portion of the vocal elements processed in the forward direction and reverse timing boundary information associated with an elapsed amount of time for a duration of the portion of the vocal elements processed in the reverse direction, and based on a comparison between the forward timing boundary information and the reverse boundary timing information, the speech decoder determines the confidence metric indicating the level of certainty for the forward timing boundary information.
28. The system of claim 27, further comprising a grammar processor for receiving text corresponding to lyrics of the audio signal, and for determining grammars corresponding to the lyrics, wherein the speech decoder performs the alignment of the grammars with corresponding textual transcriptions of the vocal elements in both the forward direction and the reverse direction by aligning the grammars of the audio signal with the grammars of the lyrics.
29. The system of claim 27, wherein the speech decoder determines a difference between the forward timing information and the reverse timing information, and based on a comparison of the difference to a predefined threshold, the speech decoder marks the portion of the vocal elements with a confidence level.
30. The system of claim 26, wherein the speech decoder synchronizes textual transcriptions of the vocal elements with the audio signal, and outputs time-annotated synchronized lyrics that indicate timing boundary information of lines of lyrics in relation to the audio signal.
PCT/US2011/036601 2010-05-18 2011-05-16 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization WO2011146366A1 (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
CN201180035459.XA CN103003875B (en) 2010-05-18 2011-05-16 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization
EP11724840A EP2572354A1 (en) 2010-05-18 2011-05-16 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization
CA2798134A CA2798134A1 (en) 2010-05-18 2011-05-16 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization
KR1020127032786A KR101413327B1 (en) 2010-05-18 2011-05-16 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US12/782,469 2010-05-18
US12/782,469 US8543395B2 (en) 2010-05-18 2010-05-18 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization

Publications (1)

Publication Number Publication Date
WO2011146366A1 true WO2011146366A1 (en) 2011-11-24

Family

ID=44310876

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2011/036601 WO2011146366A1 (en) 2010-05-18 2011-05-16 Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization

Country Status (6)

Country Link
US (1) US8543395B2 (en)
EP (1) EP2572354A1 (en)
KR (1) KR101413327B1 (en)
CN (1) CN103003875B (en)
CA (1) CA2798134A1 (en)
WO (1) WO2011146366A1 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9127013B2 (en) 2008-10-22 2015-09-08 Array Biopharma, Inc. Method of treatment using substituted pyrazolo[1,5-a] pyrimidine compounds
CN104933463A (en) * 2015-07-07 2015-09-23 杭州朗和科技有限公司 Training method of deep neural network model and equipment thereof
US9227975B2 (en) 2008-09-22 2016-01-05 Array Biopharma, Inc. Method of treatment using substituted imidazo[1,2B]pyridazine compounds
US9493476B2 (en) 2010-05-20 2016-11-15 Array Biopharma, Inc. Macrocyclic compounds as trk kinase inhibitors
US9682979B2 (en) 2009-07-09 2017-06-20 Array Biopharma, Inc. Substituted pyrazolo [1,5-A] pyrimidine compounds as TRK kinase inhibitors

Families Citing this family (68)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8781824B2 (en) * 2010-12-31 2014-07-15 Eldon Technology Limited Offline generation of subtitles
US9706247B2 (en) 2011-03-23 2017-07-11 Audible, Inc. Synchronized digital content samples
US8855797B2 (en) 2011-03-23 2014-10-07 Audible, Inc. Managing playback of synchronized content
US8948892B2 (en) 2011-03-23 2015-02-03 Audible, Inc. Managing playback of synchronized content
US9703781B2 (en) 2011-03-23 2017-07-11 Audible, Inc. Managing related digital content
US9734153B2 (en) 2011-03-23 2017-08-15 Audible, Inc. Managing related digital content
US9760920B2 (en) 2011-03-23 2017-09-12 Audible, Inc. Synchronizing digital content
US10522133B2 (en) * 2011-05-23 2019-12-31 Nuance Communications, Inc. Methods and apparatus for correcting recognition errors
US9256673B2 (en) 2011-06-10 2016-02-09 Shazam Entertainment Ltd. Methods and systems for identifying content in a data stream
JP2013025299A (en) * 2011-07-26 2013-02-04 Toshiba Corp Transcription support system and transcription support method
JP5404726B2 (en) * 2011-09-26 2014-02-05 株式会社東芝 Information processing apparatus, information processing method, and program
US9715581B1 (en) * 2011-11-04 2017-07-25 Christopher Estes Digital media reproduction and licensing
US9031493B2 (en) 2011-11-18 2015-05-12 Google Inc. Custom narration of electronic books
CN102497401A (en) * 2011-11-30 2012-06-13 上海博泰悦臻电子设备制造有限公司 Music media information acquiring method and system of vehicle-mounted music system
KR101921203B1 (en) * 2012-03-02 2018-11-22 삼성전자 주식회사 Apparatus and method for operating memo function which is associated audio recording function
US9292894B2 (en) 2012-03-14 2016-03-22 Digimarc Corporation Content recognition and synchronization using local caching
US20130268826A1 (en) * 2012-04-06 2013-10-10 Google Inc. Synchronizing progress in audio and text versions of electronic books
US9367745B2 (en) * 2012-04-24 2016-06-14 Liveclips Llc System for annotating media content for automatic content understanding
US20130283143A1 (en) * 2012-04-24 2013-10-24 Eric David Petajan System for Annotating Media Content for Automatic Content Understanding
US9275636B2 (en) 2012-05-03 2016-03-01 International Business Machines Corporation Automatic accuracy estimation for audio transcriptions
US9075760B2 (en) 2012-05-07 2015-07-07 Audible, Inc. Narration settings distribution for content customization
US9317500B2 (en) 2012-05-30 2016-04-19 Audible, Inc. Synchronizing translated digital content
US9141257B1 (en) 2012-06-18 2015-09-22 Audible, Inc. Selecting and conveying supplemental content
US8972265B1 (en) 2012-06-18 2015-03-03 Audible, Inc. Multiple voices in audio content
US9536439B1 (en) 2012-06-27 2017-01-03 Audible, Inc. Conveying questions with content
US9679608B2 (en) 2012-06-28 2017-06-13 Audible, Inc. Pacing content
US20140032537A1 (en) * 2012-07-30 2014-01-30 Ajay Shekhawat Apparatus, system, and method for music identification
US10109278B2 (en) * 2012-08-02 2018-10-23 Audible, Inc. Aligning body matter across content formats
CN103680561B (en) * 2012-08-31 2016-08-03 英业达科技有限公司 The system and method that human voice signal is synchronization with its explanatory note data
US9047356B2 (en) 2012-09-05 2015-06-02 Google Inc. Synchronizing multiple reading positions in electronic books
US9367196B1 (en) 2012-09-26 2016-06-14 Audible, Inc. Conveying branched content
US9632647B1 (en) 2012-10-09 2017-04-25 Audible, Inc. Selecting presentation positions in dynamic content
US9202520B1 (en) * 2012-10-17 2015-12-01 Amazon Technologies, Inc. Systems and methods for determining content preferences based on vocal utterances and/or movement by a user
US9223830B1 (en) 2012-10-26 2015-12-29 Audible, Inc. Content presentation analysis
US8935170B2 (en) 2012-11-27 2015-01-13 Longsand Limited Speech recognition
US9280906B2 (en) 2013-02-04 2016-03-08 Audible. Inc. Prompting a user for input during a synchronous presentation of audio content and textual content
US9472113B1 (en) 2013-02-05 2016-10-18 Audible, Inc. Synchronizing playback of digital content with physical content
US9378739B2 (en) * 2013-03-13 2016-06-28 Nuance Communications, Inc. Identifying corresponding positions in different representations of a textual work
US9058805B2 (en) * 2013-05-13 2015-06-16 Google Inc. Multiple recognizer speech recognition
CN105378830A (en) * 2013-05-31 2016-03-02 朗桑有限公司 Processing of audio data
US9317486B1 (en) 2013-06-07 2016-04-19 Audible, Inc. Synchronizing playback of digital content with captured physical content
US9489360B2 (en) 2013-09-05 2016-11-08 Audible, Inc. Identifying extra material in companion content
US9589560B1 (en) * 2013-12-19 2017-03-07 Amazon Technologies, Inc. Estimating false rejection rate in a detection system
US10776419B2 (en) 2014-05-16 2020-09-15 Gracenote Digital Ventures, Llc Audio file quality and accuracy assessment
CN104252872B (en) * 2014-09-23 2017-05-24 努比亚技术有限公司 Lyric generating method and intelligent terminal
CN105336324B (en) * 2015-11-17 2018-04-03 百度在线网络技术(北京)有限公司 A kind of Language Identification and device
CN105868318A (en) * 2016-03-25 2016-08-17 海信集团有限公司 Multimedia data type prediction method and device
KR101834854B1 (en) * 2016-07-28 2018-03-07 한국철도기술연구원 train-centric electronic interlocking system for connected train based autonomous train control system and the method thereof
US10922720B2 (en) 2017-01-11 2021-02-16 Adobe Inc. Managing content delivery via audio cues
US20180366097A1 (en) * 2017-06-14 2018-12-20 Kent E. Lovelace Method and system for automatically generating lyrics of a song
US10839826B2 (en) * 2017-08-03 2020-11-17 Spotify Ab Extracting signals from paired recordings
KR101959903B1 (en) * 2017-10-26 2019-03-19 주식회사 마이티웍스 Smart audio device
US11423208B1 (en) * 2017-11-29 2022-08-23 Amazon Technologies, Inc. Text encoding issue detection
KR102112738B1 (en) * 2017-12-06 2020-05-19 김기석 Method for displaying lyrics for karaoke device and device for the method
CN110189750B (en) * 2018-02-23 2022-11-15 株式会社东芝 Word detection system, word detection method, and recording medium
CN109102800A (en) * 2018-07-26 2018-12-28 广州酷狗计算机科技有限公司 A kind of method and apparatus that the determining lyrics show data
WO2020081872A1 (en) * 2018-10-18 2020-04-23 Warner Bros. Entertainment Inc. Characterizing content for audio-video dubbing and other transformations
US11475887B2 (en) * 2018-10-29 2022-10-18 Spotify Ab Systems and methods for aligning lyrics using a neural network
US11308943B2 (en) * 2018-10-29 2022-04-19 Spotify Ab Systems and methods for aligning lyrics using a neural network
US10785385B2 (en) * 2018-12-26 2020-09-22 NBCUniversal Media, LLC. Systems and methods for aligning text and multimedia content
US11114085B2 (en) * 2018-12-28 2021-09-07 Spotify Ab Text-to-speech from media content item snippets
EP3906552A4 (en) * 2018-12-31 2022-03-16 4S Medical Research Private Limited A method and a device for providing a performance indication to a hearing and speech impaired person learning speaking skills
KR102345625B1 (en) 2019-02-01 2021-12-31 삼성전자주식회사 Caption generation method and apparatus for performing the same
US11087738B2 (en) * 2019-06-11 2021-08-10 Lucasfilm Entertainment Company Ltd. LLC System and method for music and effects sound mix creation in audio soundtrack versioning
US11087744B2 (en) 2019-12-17 2021-08-10 Spotify Ab Masking systems and methods
CN113470662B (en) * 2020-03-31 2024-08-27 微软技术许可有限责任公司 Generating and using text-to-speech data for keyword detection system and speaker adaptation in speech recognition system
US20230386458A1 (en) * 2022-05-27 2023-11-30 Soundhound, Inc. Pre-wakeword speech processing
KR102555701B1 (en) * 2022-08-25 2023-07-17 (주)마고 Automatic synchronization between musical audio signals and lyrics using artificial intelligence

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0398574A2 (en) 1989-05-17 1990-11-22 AT&T Corp. Speech recognition employing key word modeling and non-key word modeling
US20080270138A1 (en) 2007-04-30 2008-10-30 Knight Michael J Audio content search engine
WO2009099146A1 (en) * 2008-02-05 2009-08-13 National Institute Of Advanced Industrial Science And Technology System and method for automatic time alignment of music audio signal and song lyrics
US20110005491A1 (en) 2008-01-11 2011-01-13 Toyota Jidosha Kabushiki Kaisha Fuel injection control apparatus of internal combustion engine

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5333275A (en) 1992-06-23 1994-07-26 Wheatley Barbara J System and method for time aligning speech
JP2986345B2 (en) 1993-10-18 1999-12-06 インターナショナル・ビジネス・マシーンズ・コーポレイション Voice recording indexing apparatus and method
US7117231B2 (en) * 2000-12-07 2006-10-03 International Business Machines Corporation Method and system for the automatic generation of multi-lingual synchronized sub-titles for audiovisual data
US7668718B2 (en) * 2001-07-17 2010-02-23 Custom Speech Usa, Inc. Synchronized pattern recognition source data processed by manual or automatic means for creation of shared speaker-dependent speech user profile
US7231351B1 (en) * 2002-05-10 2007-06-12 Nexidia, Inc. Transcript alignment
US7389228B2 (en) * 2002-12-16 2008-06-17 International Business Machines Corporation Speaker adaptation of vocabulary for speech recognition
US20060112812A1 (en) 2004-11-30 2006-06-01 Anand Venkataraman Method and apparatus for adapting original musical tracks for karaoke use
GB0602682D0 (en) * 2006-02-10 2006-03-22 Spinvox Ltd Spinvox speech-to-text conversion system design overview
US8005666B2 (en) 2006-10-24 2011-08-23 National Institute Of Advanced Industrial Science And Technology Automatic system for temporal alignment of music audio signal with lyrics
US20100255827A1 (en) * 2009-04-03 2010-10-07 Ubiquity Holdings On the Go Karaoke
US20100299131A1 (en) * 2009-05-21 2010-11-25 Nexidia Inc. Transcript alignment
US20100332225A1 (en) * 2009-06-29 2010-12-30 Nexidia Inc. Transcript alignment

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0398574A2 (en) 1989-05-17 1990-11-22 AT&T Corp. Speech recognition employing key word modeling and non-key word modeling
US20080270138A1 (en) 2007-04-30 2008-10-30 Knight Michael J Audio content search engine
US20110005491A1 (en) 2008-01-11 2011-01-13 Toyota Jidosha Kabushiki Kaisha Fuel injection control apparatus of internal combustion engine
WO2009099146A1 (en) * 2008-02-05 2009-08-13 National Institute Of Advanced Industrial Science And Technology System and method for automatic time alignment of music audio signal and song lyrics
US20110054910A1 (en) * 2008-02-05 2011-03-03 National Institute Of Advanced Industrial Science And Technology System and method for automatic temporal adjustment between music audio signal and lyrics

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9227975B2 (en) 2008-09-22 2016-01-05 Array Biopharma, Inc. Method of treatment using substituted imidazo[1,2B]pyridazine compounds
US9127013B2 (en) 2008-10-22 2015-09-08 Array Biopharma, Inc. Method of treatment using substituted pyrazolo[1,5-a] pyrimidine compounds
US9447104B2 (en) 2008-10-22 2016-09-20 Array Biopharma, Inc. Method of treatment using substituted pyrazolo[1,5-a]pyrimidine compounds
US9676783B2 (en) 2008-10-22 2017-06-13 Array Biopharma, Inc. Method of treatment using substituted pyrazolo[1,5-A] pyrimidine compounds
US9682979B2 (en) 2009-07-09 2017-06-20 Array Biopharma, Inc. Substituted pyrazolo [1,5-A] pyrimidine compounds as TRK kinase inhibitors
US9493476B2 (en) 2010-05-20 2016-11-15 Array Biopharma, Inc. Macrocyclic compounds as trk kinase inhibitors
CN104933463A (en) * 2015-07-07 2015-09-23 杭州朗和科技有限公司 Training method of deep neural network model and equipment thereof
CN104933463B (en) * 2015-07-07 2018-01-23 杭州朗和科技有限公司 The training method and equipment of deep neural network model

Also Published As

Publication number Publication date
KR101413327B1 (en) 2014-06-27
KR20130042493A (en) 2013-04-26
US20110288862A1 (en) 2011-11-24
CN103003875B (en) 2015-06-03
US8543395B2 (en) 2013-09-24
EP2572354A1 (en) 2013-03-27
CA2798134A1 (en) 2011-11-24
CN103003875A (en) 2013-03-27

Similar Documents

Publication Publication Date Title
US8543395B2 (en) Methods and systems for performing synchronization of audio with corresponding textual transcriptions and determining confidence values of the synchronization
Stoller et al. End-to-end lyrics alignment for polyphonic music using an audio-to-character recognition model
US8005666B2 (en) Automatic system for temporal alignment of music audio signal with lyrics
US8131545B1 (en) Aligning a transcript to audio data
Kruspe et al. Bootstrapping a System for Phoneme Recognition and Keyword Spotting in Unaccompanied Singing.
JP6578049B2 (en) Learning data generation apparatus and program thereof
US8645139B2 (en) Apparatus and method of extending pronunciation dictionary used for speech recognition
Wang et al. On the preparation and validation of a large-scale dataset of singing transcription
Dzhambazov et al. Modeling of phoneme durations for alignment between polyphonic audio and lyrics
Mesaros et al. Recognition of phonemes and words in singing
Huijbregts et al. Robust speech/non-speech classification in heterogeneous multimedia content
Davel et al. Validating smartphone-collected speech corpora
US10706867B1 (en) Global frequency-warping transformation estimation for voice timbre approximation
JP2010267012A (en) System and method for voice retrieving data
JP2017045027A (en) Speech language corpus generation device and its program
Lee et al. Word level lyrics-audio synchronization using separated vocals
Kruspe et al. Retrieval of Textual Song Lyrics from Sung Inputs.
JP5131904B2 (en) System and method for automatically associating music acoustic signal and lyrics with time
JP6027754B2 (en) Adaptation device, speech recognition device, and program thereof
JP6426971B2 (en) Learning data generation device and program thereof
WO2012150658A1 (en) Voice recognition device and voice recognition method
Byrne et al. Automatic generation of pronunciation lexicons for Mandarin spontaneous speech
Kruspe et al. Retrieval of song lyrics from sung queries
Lertwongkhanakool et al. An automatic real-time synchronization of live speech with its transcription approach
Cutugno et al. Evalita 2011: Forced alignment task

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 11724840

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2462/MUMNP/2012

Country of ref document: IN

ENP Entry into the national phase

Ref document number: 2798134

Country of ref document: CA

NENP Non-entry into the national phase

Ref country code: DE

REEP Request for entry into the european phase

Ref document number: 2011724840

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2011724840

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 20127032786

Country of ref document: KR

Kind code of ref document: A