US4908865A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speaker independent speech recognition method and system
Assignee: Texas Instruments Inc
•
Filing: 1988-12-22
•
Granted: 1990-03-13
•
Expires: 2008-12-22
Abstract
Abstract Recognition of sound units is improved by comparing frame-pair feature vectors which helps compensate for context variations in the pronunciation of sound units. A plurality of reference frames are stored of reference feature vectors representing reference words. A linear predictive coder (10) generates a plurality of spectral feature vectors for each frame of the speech signals. A filter bank system (12) transforms the spectral feature vectors to filter bank representations. A principal feature vector transformer (14) transforms the filter bank representations to an identity matrix of transformed input feature vectors. A concatenate frame system (16) concatenates the input feature vectors of adjacent frames to form the feature vector of a frame-pair. A transformer (18) and a comparator (20) compute the likelihood that each input feature vector for a frame-pair was produced by each reference frame. This computation is performed individually and independently for each reference frame-pairs. A dynamic time warper (22) constructs an optimum time path through the input speech signals for each of the computed likelihoods. A high level decision logic (24) recognizes the input speech signals as one of the reference words in response to the computed likelihoods and the optimum time paths.
Inventors
- 👤 George R. Doddington
- 👤 Enrico Bocchieri
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-27 22:58
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/12
Claims Analysis
Total Claims: 12
Claim 1
Independent Claim
145 words
A method of recognizing input speech signals organized into a sequence of frames comprising: storing a plurality of reference frames of reference feature vectors representing reference words; generating a spectral feature vector for each frame of said speech signals; concatenating said spectral feature vectors of adjacent frames to form frame-pair feature vectors; transforming said frame-pair feature vectors so that the covariance matrix of said transformed frame-pair feature vectors is an identity matrix; computing the likelihood that each transformed frame-pair feature vector was produced by each said reference frame, said computation performed individually and independently for each said reference frame; constructing an optimum time path through said input speech signals as represented by the frame-pair feature vectors for each of said computed likelihoods; and recognizing said input speech signals as one of said reference words in response to said computed likelihoods and said optimum time paths.
Claim 2
Independent Claim
145 words
A method of recognizing input speech signals organized into a sequence of frames comprising: storing a plurality of reference frames of reference feature vectors representing reference words; generating a spectral feature vector for each frame of said speech signals; concatenating said spectral feature vectors of adjacent frames to form frame-pair feature vectors; transforming said frame-pair feature vectors so that the covariance matrix of said transformed frame-pair feature vectors is an identity matrix; computing the likelihood that each transformed frame-pair feature vector was produced by each said reference frame, said computation performed individually and independently for each said reference frame; constructing an optimum time path through said input speech signals as represented by the frame-pair feature vectors for each of said computed likelihoods; and recognizing said input speech signals as one of said reference words in response to said computed likelihoods and said optimum time paths.
Claim 3
Dependent Claim
76 words
A method as set forth in claim 1, wherein the generation of a spectral feature vector includes: initially analyzing each frame of said speech signals by linear predictive coding to provide a linear predictive coding model comprising a plurality of spectral linear predictive coding parameters defining a spectral parameter vector for each frame of said speech signals; and thereafter generating said spectral feature vector from said spectral parameter vector for each frame of said speech signals.
Claim 4
Dependent Claim
65 words
A method as set forth in claim 2, wherein the generation of a spectral feature vector further includes: transforming said plurality of spectral linear predictive coding parameters to filter bank representations in defining said spectral parameter vector for each frame of said speech signals prior to the generation of said spectral feature vector from said spectral parameter vector for each frame of said speech signals.
Claim 5
Dependent Claim
47 words
The method of claim 1 wherein said step of computing the likelihood comprises: multiplying each said transformed frame-pair feature vector by a matrix representing the statistical distribution of said transformed frame-pair feature vector under the hypothesis that a particular reference frame produced each transformed frame-pair feature vector.
Claim 6
Dependent Claim
16 words
The method of claim 1 wherein said step of constructing is performed by dynamic time warping.
Claim 7
Dependent Claim
41 words
The method of claim 3 wherein said step of transforming said frame-pair feature vectors includes: estimating the covariance matrix of said frame-pair feature vectors and decomposing the resulting matrix to eigenvalues; and eliminating selected ones of said eigenvalues to reduce noise.
Claim 8
Dependent Claim
157 words
Apparatus for recognizing input speech signals organized into a sequence of frames comprising: means for storing a plurality of reference frames of reference feature vectors representing reference words; means for generating a spectral feature vector for each frame of said speech signals; means for concatenating said spectral feature vectors of adjacent frames to form frame-pair feature vectors; means for transforming said frame-pair feature vectors so that the covariance matrix of said transformed frame-pair feature vectors is an identity matrix; means for computing the likelihood that each frame-pair feature vector was produced by each said reference frame, said computation performed individually and independently for each said reference frame; means for constructing an optimum time path through said input speech signals as represented by the frame-pair feature vectors for each of said computed likelihoods; and means for recognizing said input speech signals as one of said reference words in response to said computed likelihoods and said optimum time paths.
Claim 9
Dependent Claim
157 words
Apparatus for recognizing input speech signals organized into a sequence of frames comprising: means for storing a plurality of reference frames of reference feature vectors representing reference words; means for generating a spectral feature vector for each frame of said speech signals; means for concatenating said spectral feature vectors of adjacent frames to form frame-pair feature vectors; means for transforming said frame-pair feature vectors so that the covariance matrix of said transformed frame-pair feature vectors is an identity matrix; means for computing the likelihood that each frame-pair feature vector was produced by each said reference frame, said computation performed individually and independently for each said reference frame; means for constructing an optimum time path through said input speech signals as represented by the frame-pair feature vectors for each of said computed likelihoods; and means for recognizing said input speech signals as one of said reference words in response to said computed likelihoods and said optimum time paths.
Claim 10
Dependent Claim
45 words
The apparatus of claim 7 wherein said means for transforming said frame-pair feature vectors includes: means for estimating the covariance matrix of said frame-pair feature vectors and decomposing the resulting matrix to eigenvalues; and means for eliminating selected ones of said eigenvalues to reduce noise.
Claim 11
Dependent Claim
49 words
The apparatus of claim 7 wherein said means for computing the likelihood comprises: means for multiplying each said transformed frame-pair feature vector by a matrix representing the statistical distribution of said transformed frame-pair feature vector under the hypothesis that a particular reference frame produced each transformed frame-pair feature vector.
Claim 12
Dependent Claim
15 words
The apparatus of claim 7 wherein said means for constructing comprises dynamic time warping means.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1993-03-12 | 1993-09-11 | 1994-03-13 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1997-03-12 | 1997-09-11 | 1998-03-13 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2001-03-12 | 2001-09-11 | 2002-03-13 | $7700.00 | Window Closed / Lapsed |