WO2007098316A1 - Voice recognition with speaker adaptation and registration with pitch - Google Patents
Voice recognition with speaker adaptation and registration with pitch Download PDFInfo
- Publication number
- WO2007098316A1 WO2007098316A1 PCT/US2007/061707 US2007061707W WO2007098316A1 WO 2007098316 A1 WO2007098316 A1 WO 2007098316A1 US 2007061707 W US2007061707 W US 2007061707W WO 2007098316 A1 WO2007098316 A1 WO 2007098316A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- speaker
- pitch
- voice recognition
- acoustic model
- categorization
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/065—Adaptation
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/90—Pitch determination of speech signals
Definitions
- This application relates to voice recognition and more particularly to voice recognition systems that adapt to speakers based on pitch.
- Voice and speech recognition technologies allow computers and other electronic devices equipped with a source of sound input, such as a microphone, to interpret human speech, e.g., for transcription or as an alternative method of interacting with a computer.
- Speech recognition software is being developed for use in consumer electronic devices such as mobile telephones, game platforms, personal computers and personal digital assistants.
- a time domain signal representing human speech is broken into a number of time windows and each window is converted to a frequency domain signal, e.g., by fast Fourier transform (FFT).
- FFT fast Fourier transform
- This frequency or spectral domain signal is then compressed by taking a logarithm of the spectral domain signal and then performing another FFT.
- a statistical model can be used to determine phonemes and context within the speech represented by the signal.
- the cepstrum can be seen as information about rate of change in the different spectral bands within the speech signal.
- the spectrum is usually first transformed using the Mel Frequency bands. The result is called the Mel Frequency Cepstral Coefficients or MFCCs.
- each filter function In voice recognition the spectrum is often filtered using a set of triangular-shaped filter functions.
- the filter functions divide up the spectrum into a set of partly overlapping bands that lie between a minimum frequency f mm and a maximum frequency f max .
- Each filter function is centered on a particular frequency within a frequency range of interest.
- each filter function When converted to the mel frequency scale each filter function may be expressed as a set of mel filter banks where each mel filter bank MFB 1 is given by: mf - mf mm
- MFB where the index i refers to the filter bank number and mf mm and mf m a x are the mel frequencies corresponding tof mm andf max .
- m ⁇ n andf max determines the filter banks that are used by a voice recognition algorithm.
- f mm andf max are fixed by the voice recognition model being used.
- One problem with voice recognition is that different speakers may have different vocal tract lengths and produce voice signals with correspondingly different frequency ranges.
- To compensate for this voice recognition systems may perform a vocal tract normalization of the voice signal before filtering.
- the normalization may use a function of the type:
- the components of a speech signal having N different mel frequency bands may be represented as a vector A having N components. Each component of vector A is a mel frequency coefficient of the speech signal.
- the normalization of the vector A typically involves a matrix transformation of the type:
- M m M, M NN and B is a bias vector given by:
- the matrix coefficients M 1 ⁇ and vector components B 1 are computed offline to maximize probability of an observed speech sequence in a HMM system.
- the observed probability is the computed by a Gaussian function:
- F' is a mel frequency component of the normalized speech signal. It is known that male and female speakers produce voice signals characterized by different mel frequency coefficients (MFCC).
- MFCC mel frequency coefficients
- voice recognition systems have used training to differentiate between whether the speaker is male or female and adjust the acoustic model used in voice recognition based on whether the speaker is male or female.
- the acoustic model is trained by having a number, e.g., 10, male speakers and an equal number of female speakers speak the same words to produce voice samples. Feature analyses based on the voice samples are combined together into a super model for voice recognition.
- a voice signal is obtained for an utterance of a speaker.
- a runtime pitch is determined from the voice signal for the utterance.
- the speaker is categorized based on the runtime pitch and one or more acoustic model parameters are adjusted based on a categorization of the speaker.
- a voice recognition analysis of the utterance is then performed based on the acoustic model parameters.
- FIG. 1 is a flow diagram illustrating a voice recognition algorithm according to an embodiment of the present invention.
- FIG. 2 is a block diagram illustrating a voice recognition system according to an embodiment of the present invention.
- a voice recognition method 100 may proceed as illustrated in FIG. IA.
- a voice signal is obtained for an utterance from a speaker.
- the voice signal may be obtained in any conventional fashion, e.g., using a microphone and a waveform digitizer to put the voice signal into a digital format.
- the voice signal may be obtained by over-sampling the voice signal at a sampling frequency that is greater than a working feature analysis frequency.
- the sampling frequency may be greater than a training time speech sampling rate.
- the voice signal is characterized by a working feature analysis frequency of 12 kilohertz the signal may be sampled at a sampling frequency of e.g., 16-22 kilohertz.
- a runtime pitch value p run is determined for the utterance.
- p mn may be a moving average pitch Pav g (t) may be calculated over a given time window including time t by:
- the runtime pitch p run may be related to the current pitch, e.g., by:
- Equation 2 gives:
- p run (t) may be calculated according to Equation 2 if the pitch probability is above some threshold, e.g., above about 0.4.
- the speaker categorization performed at 106 of FIG. IA may be based on the speaker's age and/or gender. For example, from training data it may be determined that average pitch for male, female and child speakers fall into different ranges. The speaker may be categorized from the pitch range into which the current pitch from the voice signal falls. By way of example, an adult male speaker has an average pitch between about 120 Hz and about 160 Hz, an adult female speaker has an average pitch between about 180 Hz and about 220 Hz and a child speaker has an average pitch greater than about 220. If the current pitch is 190 Hz, the speaker would be categorized as a female speaker. In any of these cases, the average pitch for the speaker may be included as a feature in vector F.
- the parameters of the acoustic model may be selected accordingly as indicated at 108. These parameters are then used in a voice recognition analysis at 110.
- the choice of parameters depends on the type of acoustic model used in the voice recognition analysis.
- the voice recognition analysis may filter the voice signal using a set of filter functions.
- the filter functions e.g., triangular-shaped filter functions, divide up the spectrum into a set of partly overlapping bands.
- Each voice recognition analysis uses a filter bank defined by a different maximum frequency f max and a different minimum frequency f mm .
- Thef max andf mm may be frequencies on the Hertz scale or pitches on the mel scale.
- the maximum frequency f m ⁇ x refers to an upper limit of the frequency range of the filter bank and the minimum frequency f mm refers to a lower limit of the frequency range of the filter bank.
- the values of the parameters f mm andf m ⁇ x may be adjusted dynamically at any instance of time during the voice recognition analysis, e.g., for any time window during the voice recognition analysis.
- the voice recognition analysis produces a recognition probability P r of recognition of one or more speech units.
- the speech units may be phrases, words, or sub-units of words, such as phonemes.
- the values oif mm and / ⁇ x for voice recognition analysis of the utterance may be selected accordingly. For example, if it is assumed that the speaker is a man,/ MJ may be about 70 Hz and/ mn may be about 3800 Hz. If it is assumed that the speaker is a woman, f m ⁇ x may be about 70 Hz ⁇ n ⁇ f min may be about 4200 Hz. If it is assumed that the speaker is a child, f m ⁇ x may be about 90 Hz anaf min may be about 4400 Hz.
- a recognition probability P r is from a voice analysis of the utterance based on the adjusted model parameters.
- the voice recognition analysis may use a Hidden Markov Model (HMM) to determine the units of speech in a given voice signal.
- the speech units may be words, two-word combinations or sub-word units, such as phonemes and the like.
- the HMM may be characterized by:
- M which represents the total number of Gaussians that exist in the system
- N which represents the number of distinct observable features at a given time; these features may be spectral (i.e., frequency domain) or temporal (time domain) features of the speech signal;
- A ⁇ a ⁇ , a state transition probability distribution, where each a y represents the probability that the system will transition to the j state at time t+1 if the system is initially in the i l state at time t;
- the Hidden Markov Models can be applied to the voice signal to solve one or more basic problems including: (1) the probability of a given sequence of observations obtained from the voice signal; (2) given the observation sequence, what corresponding state sequence best explains the observation sequence; and (3) how to adjust the set of model parameters A, B ⁇ to maximize the probability of a given observation sequence.
- HMMs to speech recognition is described in detail, e.g., by Lawrence Rabiner in "A tutorial on Hidden Markov Models and Selected Applications in Speech Recognition” in Proceedings of the IEEE, Vol. 77, No. 2, February 1989, which is incorporated herein by reference in its entirety for all purposes.
- the voice recognition analyses implemented at 110 may characterize speech by a number of recognizable patterns known as phonemes. Each of these phonemes can be broken down in a number of parts, e.g., a beginning, middle and ending part. It is noted that the middle part is typically the most stable since the beginning part is often affected by the preceding phoneme and the ending part is affected by the following phoneme.
- the different parts of the phonemes are characterized by frequency domain features that can be recognized by appropriate statistical analysis of the signal.
- the statistical model often uses Gaussian probability distribution functions to predict the probability for each different state of the features that make up portions of the signal that correspond to different parts of different phonemes.
- One HMM state can contain one or more Gaussians.
- a particular Gaussian for a given possible state e.g., the k th Gaussian can be represented by a set of N mean values ⁇ i ⁇ and variances O k1 .
- N mean values ⁇ i ⁇ and variances O k1 e.g., the k th Gaussian.
- the observed feature of the system may be represented as a vector having components xo...x n . These components may be spectral, cepstral, or temporal features of a given observed speech signal.
- the components xo...x n may be mel frequency cepstral coefficients (MFCCs) of the voice signal obtained at 102.
- a cepstrum is the result of taking the Fourier transform (FT) of the decibel spectrum as if it were a signal.
- the cepstrum of a time domain speech signal may be defined verbally as the Fourier transform of the log (with unwrapped phase) of the Fourier transform of the time domain signal.
- the cepstrum of a time domain signal S(t) may be represented mathematically as FT(log(FT(S(t)))+j2 ⁇ q), where q is the integer required to properly unwrap the angle or imaginary part of the complex log function.
- the cepstrum may be generated by the sequence of operations: signal - ⁇ FT ⁇ log ⁇ phase unwrapping ⁇ FT ⁇ cepstrum.
- the real cepstrum uses the logarithm function defined for real values, while the complex cepstrum uses the complex logarithm function defined for complex values also.
- the complex cepstrum holds information about magnitude and phase of the initial spectrum, allowing the reconstruction of the signal.
- the real cepstrum only uses the information of the magnitude of the spectrum.
- the voice recognition analysis implemented at 110 may use the real cepstrum.
- Certain patterns of combinations of components xo...x n correspond to units of speech (e.g., words or phrases) or sub-units, such as syllables, phonemes or other sub-units of words. Each unit or sub-unit may be regarded as a state of the system.
- the probability density function fifao... X n ) for a given Gaussian of the system (the k l Gaussian) may be any type of probability density function, e.g., a Gaussian function having the following form:
- i is an index for feature and "k” is an index for Gaussian.
- the subscript k is an index for the Gaussian function. There may be several hundred to several hundred thousand Gaussians used by the speech recognition algorithm.
- the quantity ⁇ t is a mean value for the feature X 1 in the k th Gaussian of the system.
- the quantity ⁇ t 2 is the variance for X 1 in the k th Gaussian.
- One or more Gaussians may be associated with one or more different states. For example, there may be L different states, which contain a total number of M Gaussians in the system.
- the quantity ⁇ k i is the mean for all measurements of X 1 that belong to f k (xo---XN) over all time windows of training data and O k i is the variance for the corresponding measurements used to compute ⁇ k i.
- the probability for each Gaussian can be computed equation (1) to give a corresponding recognition probability P r . From the Gaussian having the maximum probability one can build a most likely, state, word, phoneme, character, etc. for that particular time window. Note that it is also possible to use the most probable state for a given time window to help in determining the most probable state for earlier or later time windows, since these may determine a context in which the state occurs.
- a recognition method (e.g., a voice recognition method) of the type depicted in FIG. IA or FIG. IB operating as described above may be implemented as part of a signal processing apparatus 200, as depicted in FIG. 2.
- the system 200 may include a processor 201 and a memory 202 (e.g., RAM, DRAM, ROM, and the like).
- the signal processing apparatus 200 may have multiple processors 201 if parallel processing is to be implemented.
- the memory 202 includes data and code configured as described above. Specifically, the memory includes data representing signal features 204, and probability functions 206 each of which may include code, data or some combination of both code and data.
- the apparatus 200 may also include well-known support functions 210, such as input/output (I/O) elements 211, power supplies (P/S) 212, a clock (CLK) 213 and cache 214.
- the apparatus 200 may optionally include a mass storage device 215 such as a disk drive, CD- ROM drive, tape drive, or the like to store programs and/or data.
- the controller may also optionally include a display unit 216 and user interface unit 218 to facilitate interaction between the controller 200 and a user.
- the display unit 216 may be in the form of a cathode ray tube (CRT) or flat panel screen that displays text, numerals, graphical symbols or images.
- the user interface 218 may include a keyboard, mouse, joystick, light pen or other device.
- the user interface 218 may include a microphone, video camera or other signal transducing device to provide for direct capture of a signal to be analyzed.
- the processor 201, memory 202 and other components of the system 200 may exchange signals (e.g., code instructions and data) with each other via a system bus 220 as shown in FIG. 2.
- a microphone 222 may be coupled to the apparatus 200 through the I/O functions 211
- I/O generally refers to any program, operation or device that transfers data to or from the system 200 and to or from a peripheral device. Every transfer is an output from one device and an input into another.
- Peripheral devices include input-only devices, such as keyboards and mouses, output-only devices, such as printers as well as devices such as a writable CD-ROM that can act as both an input and an output device.
- peripheral device includes external devices, such as a mouse, keyboard, printer, monitor, microphone, camera, external Zip drive or scanner as well as internal devices, such as a CD-ROM drive, CD-R drive or internal modem or other peripheral such as a flash memory reader/writer, hard drive.
- the processor 201 may perform signal recognition of signal data 206 and/or probability in program code instructions of a program 204 stored and retrieved by the memory 202 and executed by the processor module 201.
- Code portions of the program 203 may conform to any one of a number of different programming languages such as Assembly, C++, JAVA or a number of other languages.
- the processor module 201 forms a general-purpose computer that becomes a specific purpose computer when executing programs such as the program code 204.
- the program code 204 is described herein as being implemented in software and executed upon a general purpose computer, those skilled in the art will realize that the method of task management could alternatively be implemented using hardware such as an application specific integrated circuit (ASIC) or other hardware circuitry.
- ASIC application specific integrated circuit
- the program code 204 may include a set of processor readable instructions that implement a method having features in common with the method 100 of FIG. IA or the method 110 of FIG. IB.
- the program 204 may generally include one or more instructions that direct the processor 201 to obtain a voice signal for an utterance of a speaker; determine a runtime pitch from the voice signal for the utterance; categorize the speaker based on the runtime pitch; adjust one or more acoustic model parameters based on a categorization of the speaker; and perform a voice recognition analysis of the utterance based on the acoustic model parameters.
- the program 204 may be part of a larger overall program, such as a program for a computer game.
- the program code 204 may prompt a speaker to speak a word or phrase (e.g., the speaker's name) during an initialization phase (e.g., at the start of a game) to provide a speech sample. From this sample, the program 204 may proceed as described above with respect to FIG. 1 to find optimal parameters (e.g.,f m ⁇ n and f max) for that speaker and run the voice recognition at 110 using those parameters. The parameters may be saved after the program concludes and used again when that speaker uses the program.
- Embodiments of the present invention provide for more robust and more accurate speech recognition.
- speech recognition employing acoustic model parameter selection using pitch-based speaker categorization with a single female speaker produced 94.8% word accuracy.
- a conventional speech recognition algorithm not employing acoustic model parameter selection using pitch-based speaker categorization achieved only 86.3% word accuracy with the same female speaker.
Abstract
Voice recognition methods and systems are disclosed. A voice signal is obtained for an utterance of a speaker. A runtime pitch is determined from the voice signal for the utterance. The speaker is categorized based on the runtime pitch and one or more acoustic model parameters are adjusted based on a categorization of the speaker. The parameter adjustment may be performed at any instance of time during the recognition. A voice recognition analysis of the utterance is then performed based on the acoustic model.
Description
VOICE RECOGNITION WITH SPEAKER ADAPTATION AND REGISTRATION
WITH PITCH
CROSS-REFERENCE TO RELATED APPLICATION
This application is related to commonly-assigned US patent application 11/358,272 entitled "VOICE RECOGNITION WITH PARALLEL GENDER AND AGE NORMALIZATION" by Ruxin Chen, which is filed the same day as the present application, the entire disclosures of which are incorporated herein by reference.
FIELD OF THE INVENTION
This application relates to voice recognition and more particularly to voice recognition systems that adapt to speakers based on pitch.
BACKGROUND OF THE INVENTION
Voice and speech recognition technologies allow computers and other electronic devices equipped with a source of sound input, such as a microphone, to interpret human speech, e.g., for transcription or as an alternative method of interacting with a computer. Speech recognition software is being developed for use in consumer electronic devices such as mobile telephones, game platforms, personal computers and personal digital assistants. In a typical speech recognition algorithm, a time domain signal representing human speech is broken into a number of time windows and each window is converted to a frequency domain signal, e.g., by fast Fourier transform (FFT). This frequency or spectral domain signal is then compressed by taking a logarithm of the spectral domain signal and then performing another FFT. From the compressed spectrum (referred to as a cepstrum), a statistical model can be used to determine phonemes and context within the speech represented by the signal. The cepstrum can be seen as information about rate of change in the different spectral bands within the speech signal. For speech recognition applications, the spectrum is usually first transformed using the Mel Frequency bands. The result is called the Mel Frequency Cepstral Coefficients or MFCCs. A frequency f in hertz (cycles per second) may be converted to a mel frequency m according to: m = (1127.01048Hz) loge(l +// 700). Similarly a mel frequency m can be converted to a frequency/in hertz using:/= (700 Hz) (e m / 1127 01048 - \y
In voice recognition the spectrum is often filtered using a set of triangular-shaped filter functions. The filter functions divide up the spectrum into a set of partly overlapping bands that lie between a minimum frequency fmm and a maximum frequency fmax. Each filter function is centered on a particular frequency within a frequency range of interest. When
converted to the mel frequency scale each filter function may be expressed as a set of mel filter banks where each mel filter bank MFB1 is given by: mf - mfmm
MFB = where the index i refers to the filter bank number and mfmm and
mfmax are the mel frequencies corresponding tofmm andfmax.
The choice offmιn andfmax determines the filter banks that are used by a voice recognition algorithm. Typically, fmm andfmax are fixed by the voice recognition model being used. One problem with voice recognition is that different speakers may have different vocal tract lengths and produce voice signals with correspondingly different frequency ranges. To compensate for this voice recognition systems may perform a vocal tract normalization of the voice signal before filtering. By way of example, the normalization may use a function of the type:
1
/'= / + sin(2^f) πarctanα U - acos(2τf) where/' is the normalized frequency and a is a parameter adjusts a curvature of the normalization function. The components of a speech signal having N different mel frequency bands may be represented as a vector A having N components. Each component of vector A is a mel frequency coefficient of the speech signal. The normalization of the vector A typically involves a matrix transformation of the type:
F' = [M]-F + B, where [M] is an NxN matrix given by:
Mn Mx, M1. M01 M,. M 2N
[M] =
Mm M, M NN and B is a bias vector given by:
B1
B0
B =
BA
F' and F are vectors of the form:
F =
where the matrix coefficients M1} and vector components B1 are computed offline to maximize probability of an observed speech sequence in a HMM system. Usually for a given frame and given feature F', the observed probability is the computed by a Gaussian function:
F' is a mel frequency component of the normalized speech signal. It is known that male and female speakers produce voice signals characterized by different mel frequency coefficients (MFCC). In the prior art, voice recognition systems have used training to differentiate between whether the speaker is male or female and adjust the acoustic model used in voice recognition based on whether the speaker is male or female. Typically, the acoustic model is trained by having a number, e.g., 10, male speakers and an equal number of female speakers speak the same words to produce voice samples. Feature analyses based on the voice samples are combined together into a super model for voice recognition.
A major drawback to the above normalization is that the vector F may have as many as 40 components. Consequently, the matrix [M] could have as many as 1600 coefficients. Computation of such a large number of coefficients can take too long for the voice recognition algorithm to adapt.
Furthermore, since prior art voice recognition systems and methods use fixed values oϊfmm, fmax, mfmm and mfmax for filtering and normalization, they do not adequately account for variations in vocal tract length amongst speakers. Consequently, speech recognition accuracy may be less than optimal. Thus, there is a need for voice recognition systems and methods that overcome such disadvantages.
SUMMARY OF THE INVENTION
The disadvantages associated with the prior art are overcome by embodiments of the present invention directed to voice recognition methods and systems. According to embodiments of the invention a voice signal is obtained for an utterance of a speaker. A runtime pitch is determined from the voice signal for the utterance. The speaker is categorized based on the runtime pitch and one or more acoustic model parameters are adjusted based on a categorization of the speaker. A voice recognition analysis of the utterance is then performed based on the acoustic model parameters.
BRIEF DESCRIPTION OF THE DRAWINGS The teachings of the present invention can be readily understood by considering the following detailed description in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow diagram illustrating a voice recognition algorithm according to an embodiment of the present invention.
FIG. 2 is a block diagram illustrating a voice recognition system according to an embodiment of the present invention.
DESCRIPTION OF THE SPECIFIC EMBODIMENTS
Although the following detailed description contains many specific details for the purposes of illustration, anyone of ordinary skill in the art will appreciate that many variations and alterations to the following details are within the scope of the invention. Accordingly, the embodiments of the invention described below are set forth without any loss of generality to, and without imposing limitations upon, the claimed invention.
According to an embodiment of the present invention a voice recognition method 100 may proceed as illustrated in FIG. IA. At 102 a voice signal is obtained for an utterance from a speaker. The voice signal may be obtained in any conventional fashion, e.g., using a microphone and a waveform digitizer to put the voice signal into a digital format. The voice signal may be obtained by over-sampling the voice signal at a sampling frequency that is greater than a working feature analysis frequency. In particular, the sampling frequency may be greater than a training time speech sampling rate. By way of example, and without limitation, if the voice signal is characterized by a working feature analysis frequency of 12 kilohertz the signal may be sampled at a sampling frequency of e.g., 16-22 kilohertz.
At 104, a runtime pitch value prun is determined for the utterance. There are a number of a ways to determine the runtime pitch prun. For example, pmn may be a moving average pitch Pavg(t) may be calculated over a given time window including time t by:
(Equation 1 ) pavg (t) = —∑p(tι ) ,
where the sum is taken over a number NP of pitch measurements taken at times U = {t-(ΝP- l),t-(NP-2),...,t} during the time window for pitch probabilities above a predetermined threshold. One simple way of computing pitch probability is correlation ( — ; — ) prob(pitch) = where correlationit) = ∑signalyt + ipignalyi) is the correlation's) , correlation of the analysis speech signal. Alternatively, the runtime pitch prun may be related to the current pitch, e.g., by:
(Equation 2) prun (t) = c - pmn (t - 1) + (1 - c ■ p(t)) , for t>0
andprun(0) =p(0), for t = 0 where c is a constant between 0 and 1 and p(t) is a current value of the pitch at time t. The value of the constant c is related to the window size. For example a value of c=0 corresponds to no window (in which case pmn(t) = p(t)) and a value of c=l corresponds to an infinite window (in which case prun(t)=pnm(t- 1)). Note that for values of t>0, pitch values for times prior to t contribute to the value of the runtime pitch />ran(t). This may be illustrated with a numerical example in which c=0.6. In such a case, Equation 2 gives:
/>™(0) = p(0) pUD = 0.6-pUO) + (1-C)T(I) = 0.6-p(0) + 0Aψ(l) prun(2) = 0.6-pUD + (l-c)7>(2) = 0.6*(0.6τ?(0) + 0.4^(1)) + 0.4^(2)
In some embodiments of the invention prun (t) may be calculated according to Equation 2 if the pitch probability is above some threshold, e.g., above about 0.4.
By way of example, the speaker categorization performed at 106 of FIG. IA may be based on the speaker's age and/or gender. For example, from training data it may be determined that average pitch for male, female and child speakers fall into different ranges. The speaker may be categorized from the pitch range into which the current pitch from the voice signal falls. By way of example, an adult male speaker has an average pitch between about 120 Hz and
about 160 Hz, an adult female speaker has an average pitch between about 180 Hz and about 220 Hz and a child speaker has an average pitch greater than about 220. If the current pitch is 190 Hz, the speaker would be categorized as a female speaker. In any of these cases, the average pitch for the speaker may be included as a feature in vector F. Once the speaker has been categorized, the parameters of the acoustic model may be selected accordingly as indicated at 108. These parameters are then used in a voice recognition analysis at 110. The choice of parameters depends on the type of acoustic model used in the voice recognition analysis. For example, the voice recognition analysis may filter the voice signal using a set of filter functions. The filter functions, e.g., triangular-shaped filter functions, divide up the spectrum into a set of partly overlapping bands. Each voice recognition analysis uses a filter bank defined by a different maximum frequency fmax and a different minimum frequency fmm. Thefmax andfmm may be frequencies on the Hertz scale or pitches on the mel scale. The maximum frequency fmαx refers to an upper limit of the frequency range of the filter bank and the minimum frequency fmm refers to a lower limit of the frequency range of the filter bank. The values of the parameters fmm andfmαx may be adjusted dynamically at any instance of time during the voice recognition analysis, e.g., for any time window during the voice recognition analysis. The voice recognition analysis produces a recognition probability Pr of recognition of one or more speech units. The speech units may be phrases, words, or sub-units of words, such as phonemes.
By way of example, once the speaker has been categorized as a male, female or child, the values oifmm and /^x for voice recognition analysis of the utterance may be selected accordingly. For example, if it is assumed that the speaker is a man,/MJ may be about 70 Hz and/mn may be about 3800 Hz. If it is assumed that the speaker is a woman, fmαx may be about 70 Hz Αnάfmin may be about 4200 Hz. If it is assumed that the speaker is a child, fmαx may be about 90 Hz anafmin may be about 4400 Hz.
At 110, a recognition probability Pr is from a voice analysis of the utterance based on the adjusted model parameters. By way of example and without loss of generality, the voice recognition analysis may use a Hidden Markov Model (HMM) to determine the units of speech in a given voice signal. The speech units may be words, two-word combinations or sub-word units, such as phonemes and the like. The HMM may be characterized by:
L, which represents a number of possible states of the system;
M, which represents the total number of Gaussians that exist in the system;
N, which represents the number of distinct observable features at a given time; these features may be spectral (i.e., frequency domain) or temporal (time domain) features of the speech signal;
A = {a^}, a state transition probability distribution, where each ay represents the probability that the system will transition to the j state at time t+1 if the system is initially in the il state at time t;
B= (bj(k)}, an observation feature probability distribution for the j state, where each b,(k) represents the probability distribution for observed values of the kth feature when the system is in the j state; and π ={ 7C1), an initial state distribution, where each component 7I1 represents the probability that the system will be in the il state at some initial time.
The Hidden Markov Models can be applied to the voice signal to solve one or more basic problems including: (1) the probability of a given sequence of observations obtained from the voice signal; (2) given the observation sequence, what corresponding state sequence best explains the observation sequence; and (3) how to adjust the set of model parameters A, B π to maximize the probability of a given observation sequence.
The application of HMMs to speech recognition is described in detail, e.g., by Lawrence Rabiner in "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition" in Proceedings of the IEEE, Vol. 77, No. 2, February 1989, which is incorporated herein by reference in its entirety for all purposes.
The voice recognition analyses implemented at 110 may characterize speech by a number of recognizable patterns known as phonemes. Each of these phonemes can be broken down in a number of parts, e.g., a beginning, middle and ending part. It is noted that the middle part is typically the most stable since the beginning part is often affected by the preceding phoneme and the ending part is affected by the following phoneme. The different parts of the phonemes are characterized by frequency domain features that can be recognized by appropriate statistical analysis of the signal. The statistical model often uses Gaussian probability distribution functions to predict the probability for each different state of the features that make up portions of the signal that correspond to different parts of different phonemes. One HMM state can contain one or more Gaussians. A particular Gaussian for a given possible state, e.g., the kth Gaussian can be represented by a set of N mean values μiα and variances Ok1. In a typical speech recognition algorithm one determines which of the
Gaussians for a given time window is the largest one. From the largest Gaussian one can infer the most probable phoneme for the time window.
By way of example, the voice recognition analysis at 110 may analyze a time domain signal to obtain N different observable signal features xo...xn, where n = N - 1. The observed feature of the system may be represented as a vector having components xo...xn. These components may be spectral, cepstral, or temporal features of a given observed speech signal.
By way of example and without limitation of the embodiments of the invention, the components xo...xn may be mel frequency cepstral coefficients (MFCCs) of the voice signal obtained at 102. A cepstrum is the result of taking the Fourier transform (FT) of the decibel spectrum as if it were a signal. The cepstrum of a time domain speech signal may be defined verbally as the Fourier transform of the log (with unwrapped phase) of the Fourier transform of the time domain signal. The cepstrum of a time domain signal S(t) may be represented mathematically as FT(log(FT(S(t)))+j2πq), where q is the integer required to properly unwrap the angle or imaginary part of the complex log function. Algorithmically: the cepstrum may be generated by the sequence of operations: signal -→ FT → log → phase unwrapping → FT → cepstrum.
There is a complex cepstrum and a real cepstrum. The real cepstrum uses the logarithm function defined for real values, while the complex cepstrum uses the complex logarithm function defined for complex values also. The complex cepstrum holds information about magnitude and phase of the initial spectrum, allowing the reconstruction of the signal. The real cepstrum only uses the information of the magnitude of the spectrum. By way of example and without loss of generality, the voice recognition analysis implemented at 110 may use the real cepstrum.
Certain patterns of combinations of components xo...xn correspond to units of speech (e.g., words or phrases) or sub-units, such as syllables, phonemes or other sub-units of words. Each unit or sub-unit may be regarded as a state of the system. The probability density function fifao... Xn) for a given Gaussian of the system (the kl Gaussian) may be any type of probability density function, e.g., a Gaussian function having the following form:
where δk = Y\ (2π * σh 2 )
i=l ...N, k=l ...M.
In the above equations, "i" is an index for feature and "k" is an index for Gaussian. In equation (1), the subscript k is an index for the Gaussian function. There may be several hundred to several hundred thousand Gaussians used by the speech recognition algorithm. The quantity μt is a mean value for the feature X1 in the kth Gaussian of the system. The quantity σt2 is the variance for X1 in the kth Gaussian. One or more Gaussians may be associated with one or more different states. For example, there may be L different states, which contain a total number of M Gaussians in the system. The quantity μki is the mean for all measurements of X1 that belong to fk(xo---XN) over all time windows of training data and Oki is the variance for the corresponding measurements used to compute μki.
The probability for each Gaussian can be computed equation (1) to give a corresponding recognition probability Pr. From the Gaussian having the maximum probability one can build a most likely, state, word, phoneme, character, etc. for that particular time window. Note that it is also possible to use the most probable state for a given time window to help in determining the most probable state for earlier or later time windows, since these may determine a context in which the state occurs.
According to embodiments of the present invention, a recognition method (e.g., a voice recognition method) of the type depicted in FIG. IA or FIG. IB operating as described above may be implemented as part of a signal processing apparatus 200, as depicted in FIG. 2. The system 200 may include a processor 201 and a memory 202 (e.g., RAM, DRAM, ROM, and the like). In addition, the signal processing apparatus 200 may have multiple processors 201 if parallel processing is to be implemented. The memory 202 includes data and code configured as described above. Specifically, the memory includes data representing signal features 204, and probability functions 206 each of which may include code, data or some combination of both code and data.
The apparatus 200 may also include well-known support functions 210, such as input/output (I/O) elements 211, power supplies (P/S) 212, a clock (CLK) 213 and cache 214. The apparatus 200 may optionally include a mass storage device 215 such as a disk drive, CD- ROM drive, tape drive, or the like to store programs and/or data. The controller may also optionally include a display unit 216 and user interface unit 218 to facilitate interaction between the controller 200 and a user. The display unit 216 may be in the form of a cathode
ray tube (CRT) or flat panel screen that displays text, numerals, graphical symbols or images. The user interface 218 may include a keyboard, mouse, joystick, light pen or other device. In addition, the user interface 218 may include a microphone, video camera or other signal transducing device to provide for direct capture of a signal to be analyzed. The processor 201, memory 202 and other components of the system 200 may exchange signals (e.g., code instructions and data) with each other via a system bus 220 as shown in FIG. 2. A microphone 222 may be coupled to the apparatus 200 through the I/O functions 211
As used herein, the term I/O generally refers to any program, operation or device that transfers data to or from the system 200 and to or from a peripheral device. Every transfer is an output from one device and an input into another. Peripheral devices include input-only devices, such as keyboards and mouses, output-only devices, such as printers as well as devices such as a writable CD-ROM that can act as both an input and an output device. The term "peripheral device" includes external devices, such as a mouse, keyboard, printer, monitor, microphone, camera, external Zip drive or scanner as well as internal devices, such as a CD-ROM drive, CD-R drive or internal modem or other peripheral such as a flash memory reader/writer, hard drive.
The processor 201 may perform signal recognition of signal data 206 and/or probability in program code instructions of a program 204 stored and retrieved by the memory 202 and executed by the processor module 201. Code portions of the program 203 may conform to any one of a number of different programming languages such as Assembly, C++, JAVA or a number of other languages. The processor module 201 forms a general-purpose computer that becomes a specific purpose computer when executing programs such as the program code 204. Although the program code 204 is described herein as being implemented in software and executed upon a general purpose computer, those skilled in the art will realize that the method of task management could alternatively be implemented using hardware such as an application specific integrated circuit (ASIC) or other hardware circuitry. As such, it should be understood that embodiments of the invention can be implemented, in whole or in part, in software, hardware or some combination of both.
In one embodiment, among others, the program code 204 may include a set of processor readable instructions that implement a method having features in common with the method 100 of FIG. IA or the method 110 of FIG. IB. The program 204 may generally include one or more instructions that direct the processor 201 to obtain a voice signal for an utterance of a speaker; determine a runtime pitch from the voice signal for the utterance; categorize the
speaker based on the runtime pitch; adjust one or more acoustic model parameters based on a categorization of the speaker; and perform a voice recognition analysis of the utterance based on the acoustic model parameters.
By way of example, the program 204 may be part of a larger overall program, such as a program for a computer game. In certain embodiments of the invention, the program code 204 may prompt a speaker to speak a word or phrase (e.g., the speaker's name) during an initialization phase (e.g., at the start of a game) to provide a speech sample. From this sample, the program 204 may proceed as described above with respect to FIG. 1 to find optimal parameters (e.g.,fmιn and f max) for that speaker and run the voice recognition at 110 using those parameters. The parameters may be saved after the program concludes and used again when that speaker uses the program.
Embodiments of the present invention provide for more robust and more accurate speech recognition. In one example of speech recognition employing acoustic model parameter selection using pitch-based speaker categorization with a single female speaker produced 94.8% word accuracy. A conventional speech recognition algorithm not employing acoustic model parameter selection using pitch-based speaker categorization achieved only 86.3% word accuracy with the same female speaker.
While the above is a complete description of the preferred embodiment of the present invention, it is possible to use various alternatives, modifications and equivalents. Therefore, the scope of the present invention should be determined not with reference to the above description but should, instead, be determined with reference to the appended claims, along with their full scope of equivalents. Any feature described herein, whether preferred or not, may be combined with any other feature described herein, whether preferred or not. In the claims that follow, the indefinite article "A", or "An" refers to a quantity of one or more of the item following the article, except where expressly stated otherwise. The appended claims are not to be interpreted as including means-plus-function limitations, unless such a limitation is explicitly recited in a given claim using the phrase "means for."
Claims
1. A method for voice recognition, the method comprising: obtaining a voice signal for an utterance of a speaker; determining a runtime pitch from the voice signal for the utterance; categorizing the speaker based on the runtime pitch; adjusting one or more acoustic model parameters based on a categorization of the speaker; and performing a voice recognition analysis of the utterance based on the acoustic model parameters.
2. The method of claim 1 wherein determining the runtime pitch includes determining a moving average pitch pavg(t) at time t given by pavg (t) = ∑ p(tt ) , where the sum is taken over a number NP of pitch
JSr i measurements taken at times t, during a time window.
3. The method of claim 2 wherein each of the pitches /?(V *s above a predetermined threshold.
4. The method of claim 2 wherein determining the runtime pitch includes a calculation of the type: P run (0 = c • P mm (t - Y) + (l - c - p(t)) , where c is a constant between 0 and 1 and p(t) is a current pitch value at time t.
5. The method of claim 1 wherein categorizing the speaker includes determining the speaker's age and/or gender.
6. The method of claim 5 wherein determining the speaker's age and/or gender includes determining whether the runtime pitch falls into a range, wherein the range depends on the speakers age and/or gender.
7. The method of claim 5 wherein determining the speaker's age and/or gender includes determining from the pitch whether the speaker is a male, female or child speaker.
8. The method of claim 1 wherein the one or more acoustic model parameters include a maximum frequency fmax and a minimum frequency fmm for a filter bank used in performing the voice recognition analysis.
9. The method of claim 8 wherein the values oifmax andfmιn are chosen based on a gender and/or an age of the speaker as determined during categorizing the speaker based on the runtime pitch.
10. The method of claim 8 wherein the values oifmax &ndfmm are chosen based whether the speaker is a male, female or child speaker during categorizing the speaker based on the runtime pitch.
11. The method of claim 8 wherein UiQfmm andfmax are adjusted dynamically at any instance of time during the recognition.
12. The method of claim 1, further comprising storing the speaker categorization and/or the one or more acoustic model parameters based on the categorization of the speaker, and associating the speaker categorization of the speaker and/or the one or more acoustic model parameters based on the categorization of the speaker with a particular speaker.
13. The method of claim 11, further comprising using the stored speaker categorization and/or the one or more acoustic model parameters based on the categorization of the speaker during a subsequent voice recognition analysis for the speaker.
14. A voice recognition system, comprising: an interface adapted to obtain a voice signal; one or more processors coupled to the interface; and a memory coupled to the interface and the processor, the memory having embodied therein a set of processor readable instructions for configured to implement a method for voice recognition, the processor readable instructions including: an instruction for obtaining a voice signal for an utterance of a speaker; an instruction for determining a runtime pitch from the voice signal for the utterance; categorizing the speaker based on the runtime pitch; an instruction for adjusting one or more acoustic model parameters based on a categorization of the speaker; and an instruction for performing a voice recognition analysis of the utterance based on the acoustic model parameters.
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2008556492A JP4959727B2 (en) | 2006-02-21 | 2007-02-06 | Speech recognition using speaker adaptation and registration by pitch |
EP07756698A EP1979894B1 (en) | 2006-02-21 | 2007-02-06 | Voice recognition with speaker adaptation based on pitch classification |
AT07756698T ATE434252T1 (en) | 2006-02-21 | 2007-02-06 | SPEECH RECOGNITION WITH SPEAKER ADAPTATION BASED ON BASE FREQUENCY CLASSIFICATION |
CN2007800061003A CN101390155B (en) | 2006-02-21 | 2007-02-06 | Voice recognition with speaker adaptation and registration with pitch |
DE602007001338T DE602007001338D1 (en) | 2006-02-21 | 2007-02-06 | SPEECH RECOGNITION WITH SPEAKER ADAPTATION BASED ON BASIC FREQUENCY CLASSIFICATION |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/358,001 | 2006-02-21 | ||
US11/358,001 US7778831B2 (en) | 2006-02-21 | 2006-02-21 | Voice recognition with dynamic filter bank adjustment based on speaker categorization determined from runtime pitch |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2007098316A1 true WO2007098316A1 (en) | 2007-08-30 |
Family
ID=38043008
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2007/061707 WO2007098316A1 (en) | 2006-02-21 | 2007-02-06 | Voice recognition with speaker adaptation and registration with pitch |
Country Status (8)
Country | Link |
---|---|
US (2) | US7778831B2 (en) |
EP (1) | EP1979894B1 (en) |
JP (1) | JP4959727B2 (en) |
CN (1) | CN101390155B (en) |
AT (1) | ATE434252T1 (en) |
DE (1) | DE602007001338D1 (en) |
ES (1) | ES2327468T3 (en) |
WO (1) | WO2007098316A1 (en) |
Families Citing this family (48)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10223934B2 (en) | 2004-09-16 | 2019-03-05 | Lena Foundation | Systems and methods for expressive language, developmental disorder, and emotion assessment, and contextual feedback |
US8938390B2 (en) | 2007-01-23 | 2015-01-20 | Lena Foundation | System and method for expressive language and developmental disorder assessment |
US8078465B2 (en) * | 2007-01-23 | 2011-12-13 | Lena Foundation | System and method for detection and analysis of speech |
US9355651B2 (en) | 2004-09-16 | 2016-05-31 | Lena Foundation | System and method for expressive language, developmental disorder, and emotion assessment |
US9240188B2 (en) | 2004-09-16 | 2016-01-19 | Lena Foundation | System and method for expressive language, developmental disorder, and emotion assessment |
US7778831B2 (en) * | 2006-02-21 | 2010-08-17 | Sony Computer Entertainment Inc. | Voice recognition with dynamic filter bank adjustment based on speaker categorization determined from runtime pitch |
CN101051464A (en) * | 2006-04-06 | 2007-10-10 | 株式会社东芝 | Registration and varification method and device identified by speaking person |
EP2126901B1 (en) | 2007-01-23 | 2015-07-01 | Infoture, Inc. | System for analysis of speech |
JP2009020291A (en) * | 2007-07-11 | 2009-01-29 | Yamaha Corp | Speech processor and communication terminal apparatus |
US20090287489A1 (en) * | 2008-05-15 | 2009-11-19 | Palm, Inc. | Speech processing for plurality of users |
US8442833B2 (en) * | 2009-02-17 | 2013-05-14 | Sony Computer Entertainment Inc. | Speech processing with source location estimation using signals from two or more microphones |
US8788256B2 (en) * | 2009-02-17 | 2014-07-22 | Sony Computer Entertainment Inc. | Multiple language voice recognition |
US8442829B2 (en) * | 2009-02-17 | 2013-05-14 | Sony Computer Entertainment Inc. | Automatic computation streaming partition for voice recognition on multiple processors with limited memory |
JP2011101110A (en) * | 2009-11-04 | 2011-05-19 | Ricoh Co Ltd | Imaging apparatus |
US8831942B1 (en) * | 2010-03-19 | 2014-09-09 | Narus, Inc. | System and method for pitch based gender identification with suspicious speaker detection |
BR112013011312A2 (en) * | 2010-11-10 | 2019-09-24 | Koninl Philips Electronics Nv | method for estimating a pattern in a signal (s) having a periodic, semiperiodic or virtually periodic component, device for estimating a pattern in a signal (s) having a periodic, semiperiodic or virtually periodic component and computer program |
US8756062B2 (en) * | 2010-12-10 | 2014-06-17 | General Motors Llc | Male acoustic model adaptation based on language-independent female speech data |
CN103282960B (en) * | 2011-01-04 | 2016-01-06 | 富士通株式会社 | Sound control apparatus, audio control method and Sound control program |
US8639508B2 (en) * | 2011-02-14 | 2014-01-28 | General Motors Llc | User-specific confidence thresholds for speech recognition |
US20120226500A1 (en) * | 2011-03-02 | 2012-09-06 | Sony Corporation | System and method for content rendering including synthetic narration |
US9224384B2 (en) | 2012-06-06 | 2015-12-29 | Cypress Semiconductor Corporation | Histogram based pre-pruning scheme for active HMMS |
US9514739B2 (en) * | 2012-06-06 | 2016-12-06 | Cypress Semiconductor Corporation | Phoneme score accelerator |
US9105268B2 (en) * | 2012-09-19 | 2015-08-11 | 24/7 Customer, Inc. | Method and apparatus for predicting intent in IVR using natural language queries |
US9319816B1 (en) * | 2012-09-26 | 2016-04-19 | Amazon Technologies, Inc. | Characterizing environment using ultrasound pilot tones |
GB2508417B (en) * | 2012-11-30 | 2017-02-08 | Toshiba Res Europe Ltd | A speech processing system |
KR20140079092A (en) * | 2012-12-18 | 2014-06-26 | 한국전자통신연구원 | Method and Apparatus for Context Independent Gender Recognition Utilizing Phoneme Transition Probability |
CN103236259B (en) * | 2013-03-22 | 2016-06-29 | 乐金电子研发中心(上海)有限公司 | Voice recognition processing and feedback system, voice replying method |
JP2015040903A (en) * | 2013-08-20 | 2015-03-02 | ソニー株式会社 | Voice processor, voice processing method and program |
US20150154002A1 (en) * | 2013-12-04 | 2015-06-04 | Google Inc. | User interface customization based on speaker characteristics |
CN103714812A (en) * | 2013-12-23 | 2014-04-09 | 百度在线网络技术(北京)有限公司 | Voice identification method and voice identification device |
CN110459214B (en) * | 2014-01-20 | 2022-05-13 | 华为技术有限公司 | Voice interaction method and device |
US9412358B2 (en) | 2014-05-13 | 2016-08-09 | At&T Intellectual Property I, L.P. | System and method for data-driven socially customized models for language generation |
US10127927B2 (en) | 2014-07-28 | 2018-11-13 | Sony Interactive Entertainment Inc. | Emotional speech processing |
CN105895078A (en) * | 2015-11-26 | 2016-08-24 | 乐视致新电子科技(天津)有限公司 | Speech recognition method used for dynamically selecting speech model and device |
US11455985B2 (en) * | 2016-04-26 | 2022-09-27 | Sony Interactive Entertainment Inc. | Information processing apparatus |
CN105895105B (en) * | 2016-06-06 | 2020-05-05 | 北京云知声信息技术有限公司 | Voice processing method and device |
US9818406B1 (en) | 2016-06-23 | 2017-11-14 | Intuit Inc. | Adjusting user experience based on paralinguistic information |
US10135989B1 (en) | 2016-10-27 | 2018-11-20 | Intuit Inc. | Personalized support routing based on paralinguistic information |
US10515632B2 (en) | 2016-11-15 | 2019-12-24 | At&T Intellectual Property I, L.P. | Asynchronous virtual assistant |
US10431236B2 (en) * | 2016-11-15 | 2019-10-01 | Sphero, Inc. | Dynamic pitch adjustment of inbound audio to improve speech recognition |
US10803857B2 (en) * | 2017-03-10 | 2020-10-13 | James Jordan Rosenberg | System and method for relative enhancement of vocal utterances in an acoustically cluttered environment |
US10468032B2 (en) * | 2017-04-10 | 2019-11-05 | Intel Corporation | Method and system of speaker recognition using context aware confidence modeling |
US10331402B1 (en) * | 2017-05-30 | 2019-06-25 | Amazon Technologies, Inc. | Search and knowledge base question answering for a voice user interface |
WO2019113477A1 (en) | 2017-12-07 | 2019-06-13 | Lena Foundation | Systems and methods for automatic determination of infant cry and discrimination of cry from fussiness |
US10818296B2 (en) | 2018-06-21 | 2020-10-27 | Intel Corporation | Method and system of robust speaker recognition activation |
KR20210001529A (en) * | 2019-06-28 | 2021-01-06 | 엘지전자 주식회사 | Robot, server connected thereto, and method for recognizing voice using robot |
CN110808052A (en) * | 2019-11-12 | 2020-02-18 | 深圳市瑞讯云技术有限公司 | Voice recognition method and device and electronic equipment |
US11664033B2 (en) * | 2020-06-15 | 2023-05-30 | Samsung Electronics Co., Ltd. | Electronic apparatus and controlling method thereof |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0866442A2 (en) * | 1997-03-20 | 1998-09-23 | AT&T Corp. | Combining frequency warping and spectral shaping in HMM based speech recognition |
Family Cites Families (130)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
USRE33597E (en) | 1982-10-15 | 1991-05-28 | Hidden Markov model speech recognition arrangement | |
JPS6075898A (en) * | 1983-09-30 | 1985-04-30 | 三菱電機株式会社 | Word voice recognition equipment |
JPH0646359B2 (en) * | 1984-02-10 | 1994-06-15 | 三菱電機株式会社 | Word speech recognizer |
US4956865A (en) | 1985-01-30 | 1990-09-11 | Northern Telecom Limited | Speech recognition |
JPH01102599A (en) | 1987-10-12 | 1989-04-20 | Internatl Business Mach Corp <Ibm> | Voice recognition |
US5129002A (en) | 1987-12-16 | 1992-07-07 | Matsushita Electric Industrial Co., Ltd. | Pattern recognition apparatus |
JPH0293597A (en) | 1988-09-30 | 1990-04-04 | Nippon I B M Kk | Speech recognition device |
JPH02273798A (en) * | 1989-04-14 | 1990-11-08 | Sekisui Chem Co Ltd | Speaker recognition system |
US5228087A (en) | 1989-04-12 | 1993-07-13 | Smiths Industries Public Limited Company | Speech recognition apparatus and methods |
US4977598A (en) | 1989-04-13 | 1990-12-11 | Texas Instruments Incorporated | Efficient pruning algorithm for hidden markov model speech recognition |
US5509104A (en) | 1989-05-17 | 1996-04-16 | At&T Corp. | Speech recognition employing key word modeling and non-key word modeling |
CA2015410C (en) | 1989-05-17 | 1996-04-02 | Chin H. Lee | Speech recognition employing key word modeling and non-key word modeling |
US5148489A (en) | 1990-02-28 | 1992-09-15 | Sri International | Method for spectral estimation to improve noise robustness for speech recognition |
US5794190A (en) | 1990-04-26 | 1998-08-11 | British Telecommunications Public Limited Company | Speech pattern recognition using pattern recognizers and classifiers |
US5345536A (en) | 1990-12-21 | 1994-09-06 | Matsushita Electric Industrial Co., Ltd. | Method of speech recognition |
US5268990A (en) | 1991-01-31 | 1993-12-07 | Sri International | Method for recognizing speech using linguistically-motivated hidden Markov models |
JP3050934B2 (en) | 1991-03-22 | 2000-06-12 | 株式会社東芝 | Voice recognition method |
US5222190A (en) | 1991-06-11 | 1993-06-22 | Texas Instruments Incorporated | Apparatus and method for identifying a speech pattern |
JP2662120B2 (en) | 1991-10-01 | 1997-10-08 | 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 | Speech recognition device and processing unit for speech recognition |
US5502790A (en) | 1991-12-24 | 1996-03-26 | Oki Electric Industry Co., Ltd. | Speech recognition method and system using triphones, diphones, and phonemes |
JPH05257492A (en) | 1992-03-13 | 1993-10-08 | Toshiba Corp | Voice recognizing system |
JPH0782348B2 (en) | 1992-03-21 | 1995-09-06 | 株式会社エイ・ティ・アール自動翻訳電話研究所 | Subword model generation method for speech recognition |
JP2795058B2 (en) | 1992-06-03 | 1998-09-10 | 松下電器産業株式会社 | Time series signal processing device |
US5455888A (en) | 1992-12-04 | 1995-10-03 | Northern Telecom Limited | Speech bandwidth extension method and apparatus |
JP3272842B2 (en) | 1992-12-17 | 2002-04-08 | ゼロックス・コーポレーション | Processor-based decision method |
US5438630A (en) | 1992-12-17 | 1995-08-01 | Xerox Corporation | Word spotting in bitmap images using word bounding boxes and hidden Markov models |
US5535305A (en) | 1992-12-31 | 1996-07-09 | Apple Computer, Inc. | Sub-partitioned vector quantization of probability density functions |
US5473728A (en) | 1993-02-24 | 1995-12-05 | The United States Of America As Represented By The Secretary Of The Navy | Training of homoscedastic hidden Markov models for automatic speech recognition |
US5459798A (en) | 1993-03-19 | 1995-10-17 | Intel Corporation | System and method of pattern recognition employing a multiprocessing pipelined apparatus with private pattern memory |
JPH0728487A (en) | 1993-03-26 | 1995-01-31 | Texas Instr Inc <Ti> | Voice recognition |
EP0708958B1 (en) | 1993-07-13 | 2001-04-11 | Theodore Austin Bordeaux | Multi-language speech recognition system |
US5627939A (en) | 1993-09-03 | 1997-05-06 | Microsoft Corporation | Speech recognition system and method employing data compression |
WO1995009416A1 (en) | 1993-09-30 | 1995-04-06 | Apple Computer, Inc. | Continuous reference adaptation in a pattern recognition system |
US5615296A (en) | 1993-11-12 | 1997-03-25 | International Business Machines Corporation | Continuous speech recognition and voice response system and method to enable conversational dialogues with microprocessors |
ZA948426B (en) | 1993-12-22 | 1995-06-30 | Qualcomm Inc | Distributed voice recognition system |
JP2737624B2 (en) | 1993-12-27 | 1998-04-08 | 日本電気株式会社 | Voice recognition device |
FI98162C (en) | 1994-05-30 | 1997-04-25 | Tecnomen Oy | Speech recognition method based on HMM model |
US6061652A (en) | 1994-06-13 | 2000-05-09 | Matsushita Electric Industrial Co., Ltd. | Speech recognition apparatus |
US5825978A (en) | 1994-07-18 | 1998-10-20 | Sri International | Method and apparatus for speech recognition using optimized partial mixture tying of HMM state functions |
US5602960A (en) | 1994-09-30 | 1997-02-11 | Apple Computer, Inc. | Continuous mandarin chinese speech recognition system having an integrated tone classifier |
JP3581401B2 (en) | 1994-10-07 | 2004-10-27 | キヤノン株式会社 | Voice recognition method |
US5680506A (en) | 1994-12-29 | 1997-10-21 | Lucent Technologies Inc. | Apparatus and method for speech signal analysis |
DE19501599C1 (en) | 1995-01-20 | 1996-05-02 | Daimler Benz Ag | Speech recognition method for word sequence |
US5680510A (en) | 1995-01-26 | 1997-10-21 | Apple Computer, Inc. | System and method for generating and using context dependent sub-syllable models to recognize a tonal language |
US5751905A (en) * | 1995-03-15 | 1998-05-12 | International Business Machines Corporation | Statistical acoustic processing method and apparatus for speech recognition using a toned phoneme system |
US5617509A (en) | 1995-03-29 | 1997-04-01 | Motorola, Inc. | Method, apparatus, and radio optimizing Hidden Markov Model speech recognition |
US5719996A (en) | 1995-06-30 | 1998-02-17 | Motorola, Inc. | Speech recognition in selective call systems |
DE69616568T2 (en) | 1995-08-24 | 2002-07-11 | British Telecomm | PATTERN RECOGNITION |
JPH0981183A (en) | 1995-09-14 | 1997-03-28 | Pioneer Electron Corp | Generating method for voice model and voice recognition device using the method |
GB2305288A (en) | 1995-09-15 | 1997-04-02 | Ibm | Speech recognition system |
US6067520A (en) | 1995-12-29 | 2000-05-23 | Lee And Li | System and method of recognizing continuous mandarin speech utilizing chinese hidden markou models |
GB9602691D0 (en) | 1996-02-09 | 1996-04-10 | Canon Kk | Word model generation |
US5696873A (en) * | 1996-03-18 | 1997-12-09 | Advanced Micro Devices, Inc. | Vocoder system and method for performing pitch estimation using an adaptive correlation sample window |
US5880788A (en) | 1996-03-25 | 1999-03-09 | Interval Research Corporation | Automated synchronization of video image sequences to new soundtracks |
US5913193A (en) * | 1996-04-30 | 1999-06-15 | Microsoft Corporation | Method and system of runtime acoustic unit selection for speech synthesis |
US5937384A (en) | 1996-05-01 | 1999-08-10 | Microsoft Corporation | Method and system for speech recognition using continuous density hidden Markov models |
US5860062A (en) | 1996-06-21 | 1999-01-12 | Matsushita Electric Industrial Co., Ltd. | Speech recognition apparatus and speech recognition method |
US5963903A (en) | 1996-06-28 | 1999-10-05 | Microsoft Corporation | Method and system for dynamically adjusted training for speech recognition |
JP3302266B2 (en) | 1996-07-23 | 2002-07-15 | 沖電気工業株式会社 | Learning Hidden Markov Model |
US5835890A (en) | 1996-08-02 | 1998-11-10 | Nippon Telegraph And Telephone Corporation | Method for speaker adaptation of speech models recognition scheme using the method and recording medium having the speech recognition method recorded thereon |
JPH10149187A (en) * | 1996-11-19 | 1998-06-02 | Yamaha Corp | Audio information extracting device |
JP3501199B2 (en) * | 1997-02-17 | 2004-03-02 | 日本電信電話株式会社 | Acoustic signal separation method |
GB9706174D0 (en) | 1997-03-25 | 1997-11-19 | Secr Defence | Recognition system |
JP3033514B2 (en) | 1997-03-31 | 2000-04-17 | 日本電気株式会社 | Large vocabulary speech recognition method and apparatus |
US5893059A (en) | 1997-04-17 | 1999-04-06 | Nynex Science And Technology, Inc. | Speech recoginition methods and apparatus |
US6456965B1 (en) * | 1997-05-20 | 2002-09-24 | Texas Instruments Incorporated | Multi-stage pitch and mixed voicing estimation for harmonic speech coders |
US5963906A (en) | 1997-05-20 | 1999-10-05 | At & T Corp | Speech recognition training |
US6032116A (en) | 1997-06-27 | 2000-02-29 | Advanced Micro Devices, Inc. | Distance measure in a speech recognition system for speech recognition using frequency shifting factors to compensate for input signal frequency shifts |
US6009390A (en) | 1997-09-11 | 1999-12-28 | Lucent Technologies Inc. | Technique for selective use of Gaussian kernels and mixture component weights of tied-mixture hidden Markov models for speech recognition |
US6151573A (en) | 1997-09-17 | 2000-11-21 | Texas Instruments Incorporated | Source normalization training for HMM modeling of speech |
FR2769117B1 (en) | 1997-09-29 | 2000-11-10 | Matra Comm | LEARNING METHOD IN A SPEECH RECOGNITION SYSTEM |
FR2769118B1 (en) | 1997-09-29 | 1999-12-03 | Matra Communication | SPEECH RECOGNITION PROCESS |
US5983180A (en) | 1997-10-23 | 1999-11-09 | Softsound Limited | Recognition of sequential data using finite state sequence models organized in a tree structure |
US6188982B1 (en) | 1997-12-01 | 2001-02-13 | Industrial Technology Research Institute | On-line background noise adaptation of parallel model combination HMM with discriminative learning using weighted HMM for noisy speech recognition |
US6151574A (en) | 1997-12-05 | 2000-11-21 | Lucent Technologies Inc. | Technique for adaptation of hidden markov models for speech recognition |
JP2965537B2 (en) | 1997-12-10 | 1999-10-18 | 株式会社エイ・ティ・アール音声翻訳通信研究所 | Speaker clustering processing device and speech recognition device |
US6226612B1 (en) | 1998-01-30 | 2001-05-01 | Motorola, Inc. | Method of evaluating an utterance in a speech recognition system |
US6148284A (en) | 1998-02-23 | 2000-11-14 | At&T Corporation | Method and apparatus for automatic speech recognition using Markov processes on curves |
JP3412496B2 (en) | 1998-02-25 | 2003-06-03 | 三菱電機株式会社 | Speaker adaptation device and speech recognition device |
US6112175A (en) | 1998-03-02 | 2000-08-29 | Lucent Technologies Inc. | Speaker adaptation using discriminative linear regression on time-varying mean parameters in trended HMM |
JP2986792B2 (en) | 1998-03-16 | 1999-12-06 | 株式会社エイ・ティ・アール音声翻訳通信研究所 | Speaker normalization processing device and speech recognition device |
EP1078355B1 (en) | 1998-05-11 | 2002-10-02 | Siemens Aktiengesellschaft | Method and array for introducing temporal correlation in hidden markov models for speech recognition |
EP1084490B1 (en) | 1998-05-11 | 2003-03-26 | Siemens Aktiengesellschaft | Arrangement and method for computer recognition of a predefined vocabulary in spoken language |
JP3156668B2 (en) | 1998-06-19 | 2001-04-16 | 日本電気株式会社 | Voice recognition device |
US6980952B1 (en) | 1998-08-15 | 2005-12-27 | Texas Instruments Incorporated | Source normalization training for HMM modeling of speech |
US6138095A (en) | 1998-09-03 | 2000-10-24 | Lucent Technologies Inc. | Speech recognition |
JP3000999B1 (en) | 1998-09-08 | 2000-01-17 | セイコーエプソン株式会社 | Speech recognition method, speech recognition device, and recording medium recording speech recognition processing program |
EP1126438B1 (en) | 1998-09-09 | 2008-07-16 | Asahi Kasei Kabushiki Kaisha | Speech recognizer and speech recognition method |
US6766288B1 (en) * | 1998-10-29 | 2004-07-20 | Paul Reed Smith Guitars | Fast find fundamental method |
US6292776B1 (en) | 1999-03-12 | 2001-09-18 | Lucent Technologies Inc. | Hierarchial subband linear predictive cepstral features for HMM-based speech recognition |
GB2348035B (en) | 1999-03-19 | 2003-05-28 | Ibm | Speech recognition system |
US6526380B1 (en) | 1999-03-26 | 2003-02-25 | Koninklijke Philips Electronics N.V. | Speech recognition system having parallel large vocabulary recognition engines |
US7058573B1 (en) * | 1999-04-20 | 2006-06-06 | Nuance Communications Inc. | Speech recognition system to selectively utilize different speech recognition techniques over multiple speech recognition passes |
US6405168B1 (en) | 1999-09-30 | 2002-06-11 | Conexant Systems, Inc. | Speaker dependent speech recognition training using simplified hidden markov modeling and robust end-point detection |
JP3632529B2 (en) | 1999-10-26 | 2005-03-23 | 日本電気株式会社 | Voice recognition apparatus and method, and recording medium |
KR100531549B1 (en) * | 1999-10-29 | 2005-11-28 | 마쯔시다덴기산교 가부시키가이샤 | Device for normalizing voice pitch for voice recognition |
US6442519B1 (en) * | 1999-11-10 | 2002-08-27 | International Business Machines Corp. | Speaker model adaptation via network of similar users |
JP2003514260A (en) * | 1999-11-11 | 2003-04-15 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | Tone features for speech recognition |
JP3814459B2 (en) | 2000-03-31 | 2006-08-30 | キヤノン株式会社 | Speech recognition method and apparatus, and storage medium |
US6629073B1 (en) | 2000-04-27 | 2003-09-30 | Microsoft Corporation | Speech recognition method and apparatus utilizing multi-unit models |
US7272561B2 (en) * | 2000-07-13 | 2007-09-18 | Asahi Kasei Kabushiki Kaisha | Speech recognition device and speech recognition method |
US6671669B1 (en) | 2000-07-18 | 2003-12-30 | Qualcomm Incorporated | combined engine system and method for voice recognition |
TW473704B (en) | 2000-08-30 | 2002-01-21 | Ind Tech Res Inst | Adaptive voice recognition method with noise compensation |
DE10043946C2 (en) | 2000-09-06 | 2002-12-12 | Siemens Ag | Compress HMM prototypes |
JP3932789B2 (en) | 2000-09-20 | 2007-06-20 | セイコーエプソン株式会社 | HMM output probability calculation method and speech recognition apparatus |
US6681207B2 (en) | 2001-01-12 | 2004-01-20 | Qualcomm Incorporated | System and method for lossy compression of voice recognition models |
US20020169604A1 (en) | 2001-03-09 | 2002-11-14 | Damiba Bertrand A. | System, method and computer program product for genre-based grammars and acoustic models in a speech recognition framework |
JP2002366187A (en) | 2001-06-08 | 2002-12-20 | Sony Corp | Device and method for recognizing voice, program and recording medium |
US6701293B2 (en) | 2001-06-13 | 2004-03-02 | Intel Corporation | Combining N-best lists from multiple speech recognizers |
US6493668B1 (en) * | 2001-06-15 | 2002-12-10 | Yigal Brandman | Speech feature extraction system |
JP2003066991A (en) * | 2001-08-22 | 2003-03-05 | Seiko Epson Corp | Method and apparatus for outputting voice recognition result and recording medium with program for outputting and processing voice recognition result recorded thereon |
CA2359544A1 (en) | 2001-10-22 | 2003-04-22 | Dspfactory Ltd. | Low-resource real-time speech recognition system using an oversampled filterbank |
US6721699B2 (en) * | 2001-11-12 | 2004-04-13 | Intel Corporation | Method and system of Chinese speech pitch extraction |
US20030220788A1 (en) * | 2001-12-17 | 2003-11-27 | Xl8 Systems, Inc. | System and method for speech recognition and transcription |
JP2005227794A (en) * | 2002-11-21 | 2005-08-25 | Matsushita Electric Ind Co Ltd | Device and method for creating standard model |
US7133535B2 (en) | 2002-12-21 | 2006-11-07 | Microsoft Corp. | System and method for real time lip synchronization |
JP2004297273A (en) * | 2003-03-26 | 2004-10-21 | Kenwood Corp | Apparatus and method for eliminating noise in sound signal, and program |
US7389230B1 (en) * | 2003-04-22 | 2008-06-17 | International Business Machines Corporation | System and method for classification of voice signals |
US7499857B2 (en) * | 2003-05-15 | 2009-03-03 | Microsoft Corporation | Adaptation of compressed acoustic models |
KR100511248B1 (en) * | 2003-06-13 | 2005-08-31 | 홍광석 | An Amplitude Warping Approach to Intra-Speaker Normalization for Speech Recognition |
US7328154B2 (en) | 2003-08-13 | 2008-02-05 | Matsushita Electrical Industrial Co., Ltd. | Bubble splitting for compact acoustic modeling |
US20050065789A1 (en) | 2003-09-23 | 2005-03-24 | Sherif Yacoub | System and method with automated speech recognition engines |
JP2005164988A (en) * | 2003-12-03 | 2005-06-23 | Xanavi Informatics Corp | Frequency switching device and information processing apparatus |
JP2005173008A (en) * | 2003-12-09 | 2005-06-30 | Canon Inc | Voice analysis processing, voice processor using same, and medium |
JP2005215888A (en) * | 2004-01-28 | 2005-08-11 | Yasunori Kobori | Display device for text sentence |
WO2005088607A1 (en) | 2004-03-12 | 2005-09-22 | Siemens Aktiengesellschaft | User and vocabulary-adaptive determination of confidence and rejecting thresholds |
US7844045B2 (en) | 2004-06-16 | 2010-11-30 | Panasonic Corporation | Intelligent call routing and call supervision method for call centers |
KR100655491B1 (en) | 2004-12-21 | 2006-12-11 | 한국전자통신연구원 | Two stage utterance verification method and device of speech recognition system |
US7970613B2 (en) | 2005-11-12 | 2011-06-28 | Sony Computer Entertainment Inc. | Method and system for Gaussian probability data bit reduction and computation |
US7778831B2 (en) | 2006-02-21 | 2010-08-17 | Sony Computer Entertainment Inc. | Voice recognition with dynamic filter bank adjustment based on speaker categorization determined from runtime pitch |
-
2006
- 2006-02-21 US US11/358,001 patent/US7778831B2/en active Active
-
2007
- 2007-02-06 DE DE602007001338T patent/DE602007001338D1/en active Active
- 2007-02-06 CN CN2007800061003A patent/CN101390155B/en active Active
- 2007-02-06 ES ES07756698T patent/ES2327468T3/en active Active
- 2007-02-06 EP EP07756698A patent/EP1979894B1/en active Active
- 2007-02-06 JP JP2008556492A patent/JP4959727B2/en active Active
- 2007-02-06 WO PCT/US2007/061707 patent/WO2007098316A1/en active Application Filing
- 2007-02-06 AT AT07756698T patent/ATE434252T1/en not_active IP Right Cessation
-
2010
- 2010-07-21 US US12/841,101 patent/US8050922B2/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0866442A2 (en) * | 1997-03-20 | 1998-09-23 | AT&T Corp. | Combining frequency warping and spectral shaping in HMM based speech recognition |
Non-Patent Citations (2)
Title |
---|
LI LEE ET AL: "Speaker normalization using efficient frequency warping procedures", 1996 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING - PROCEEDINGS. (ICASSP). ATLANTA, MAY 7 - 10, 1996, IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING - PROCEEDINGS. (ICASSP), NEW YORK, IEEE, US, vol. VOL. 1 CONF. 21, 1996, pages 353 - 356, XP002093540, ISBN: 0-7803-3193-1 * |
SINHA R ET AL: "Non-uniform scaling based speaker normalization", 2002 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING. PROCEEDINGS. (ICASSP). ORLANDO, FL, MAY 13 - 17, 2002, IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING (ICASSP), NEW YORK, NY : IEEE, US, vol. VOL. 4 OF 4, 13 May 2002 (2002-05-13), pages I - 589, XP010804772, ISBN: 0-7803-7402-9 * |
Also Published As
Publication number | Publication date |
---|---|
ES2327468T3 (en) | 2009-10-29 |
JP4959727B2 (en) | 2012-06-27 |
EP1979894B1 (en) | 2009-06-17 |
CN101390155A (en) | 2009-03-18 |
ATE434252T1 (en) | 2009-07-15 |
US20070198263A1 (en) | 2007-08-23 |
CN101390155B (en) | 2012-08-15 |
US7778831B2 (en) | 2010-08-17 |
EP1979894A1 (en) | 2008-10-15 |
US8050922B2 (en) | 2011-11-01 |
DE602007001338D1 (en) | 2009-07-30 |
JP2009527801A (en) | 2009-07-30 |
US20100324898A1 (en) | 2010-12-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7778831B2 (en) | Voice recognition with dynamic filter bank adjustment based on speaker categorization determined from runtime pitch | |
US8010358B2 (en) | Voice recognition with parallel gender and age normalization | |
US7970613B2 (en) | Method and system for Gaussian probability data bit reduction and computation | |
US9792897B1 (en) | Phoneme-expert assisted speech recognition and re-synthesis | |
US7689419B2 (en) | Updating hidden conditional random field model parameters after processing individual training samples | |
US20030200086A1 (en) | Speech recognition apparatus, speech recognition method, and computer-readable recording medium in which speech recognition program is recorded | |
US20030088412A1 (en) | Pattern recognition using an observable operator model | |
CN109036381A (en) | Method of speech processing and device, computer installation and readable storage medium storing program for executing | |
US20030200090A1 (en) | Speech recognition apparatus, speech recognition method, and computer-readable recording medium in which speech recognition program is recorded | |
JP2009139894A (en) | Noise suppressing device, speech recognition device, noise suppressing method and program | |
Eringis et al. | Improving speech recognition rate through analysis parameters | |
JP2009003008A (en) | Noise-suppressing device, speech recognition device, noise-suppressing method and program | |
Guo et al. | Robust speaker identification via fusion of subglottal resonances and cepstral features | |
Yuan et al. | Speech recognition on DSP: issues on computational efficiency and performance analysis | |
Saloni et al. | Disease detection using voice analysis: A review | |
Akafi et al. | Assessment of hypernasality for children with cleft palate based on cepstrum analysis | |
Yavuz et al. | A Phoneme-Based Approach for Eliminating Out-of-vocabulary Problem Turkish Speech Recognition Using Hidden Markov Model. | |
Selvan et al. | Speaker recognition system for security applications | |
US9928832B2 (en) | Method and apparatus for classifying lexical stress | |
Karpov | Efficient speaker recognition for mobile devices | |
Laleye et al. | Automatic boundary detection based on entropy measures for text-independent syllable segmentation | |
EP1369847B1 (en) | Speech recognition method and system | |
Tripathi et al. | Robust vowel region detection method for multimode speech | |
Bhable et al. | Automatic speech recognition (ASR) of isolated words in Hindi low resource language | |
Sherly et al. | ASR Models from Conventional Statistical Models to Transformers and Transfer Learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
WWE | Wipo information: entry into national phase |
Ref document number: 2007756698 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 200780006100.3 Country of ref document: CN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2008556492 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |