Patent US5604839A successfully refreshed from Google Patents.
US5604839A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method and system for improving speech recognition through front-end normalization of feature vectors
Assignee: Microsoft Technology Licensing LLC
•
Filing: 1994-07-29
•
Granted: 1997-02-18
•
Expires: 2014-07-29
Abstract
Abstract A method and system for improving speech recognition through front-end normalization of feature vectors are provided. Speech to be recognized is spoken into a microphone, amplified by an amplifier, and converted from an analog signal to a digital signal by an analog-to-digital ("A/D") converter. The digital signal from the A/D converter is input to a feature extractor that breaks down the signal into frames of speech and then extracts a feature vector from each of the frames. The feature vector is input to an input normalizer that normalizes the vector. The input normalizer normalizes the feature vector by computing a correction vector and subtracting the correction vector from the feature vector. The correction vector is computed based on the probability of the current frame of speech being noise and based on the average noise and speech feature vectors for a current utterance and a database of utterances. The normalization of the feature vector reduces the effect of changes in the acoustical environment on the feature vector. The normalized feature vector is input to a pattern matcher that compares the normalized vector to feature models stored in the database to find an exact match or a best match.
Inventors
- 👤 Alejandro Acero
- 👤 Xuedong Huang
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 14:44
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/02; G10L15/08; G10L15/10; G10L15/12
Claims Analysis
Total Claims: 25
Claim 1
Independent Claim
150 words
A method for improving speech recognition through front-end normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the method comprising the steps of: providing a database of known utterances, the database of utterances having an average noise feature vector and an average speech feature vector; receiving a feature vector representing a frame of speech in an utterance to be recognized, the frame of speech having a probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; computing a correction vector based on the probability of the frame of speech being noise and based on the average noise and speech feature vectors for the utterance and the database of utterances; and computing a normalized feature vector based on the feature vector and the correction vector.
Claim 2
Independent Claim
150 words
A method for improving speech recognition through front-end normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the method comprising the steps of: providing a database of known utterances, the database of utterances having an average noise feature vector and an average speech feature vector; receiving a feature vector representing a frame of speech in an utterance to be recognized, the frame of speech having a probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; computing a correction vector based on the probability of the frame of speech being noise and based on the average noise and speech feature vectors for the utterance and the database of utterances; and computing a normalized feature vector based on the feature vector and the correction vector.
Claim 3
Dependent Claim
21 words
The method of claim 1, wherein the step of receiving a feature vector comprises the step of receiving a cepstral vector.
Claim 4
Dependent Claim
33 words
The method of claim 1, wherein the probability of the frame of speech being noise and the average noise and speech feature vectors for the utterance are updated for each frame of speech.
Claim 5
Dependent Claim
115 words
The method of claim 1, wherein the step of computing a correction vector includes the steps of: computing the probability of the frame of speech being noise based on a distortion measure of the frame of speech; computing the average noise and speech feature vectors for the utterance; computing the average noise and speech feature vectors for the database of utterances; and computing the correction vector based on the probability of the frame of speech being noise, the difference between the average noise for the utterance and the average noise for the database of utterances, and the difference between the speech vectors for the utterance and the average speech vectors for the database of utterances.
Claim 6
Independent Claim
183 words
A method for improving speech recognition through front-end normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the method comprising the steps of: providing a database of known utterances, the database of utterances having an average noise feature vector and an average speech feature vector; receiving a feature vector x j representing a frame of speech j in an utterance to be recognized, the frame of speech having an a posteriori probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; computing a correction vector r(x j ) as: r(x.sub.j)=p.sub.j (n.sub.j-1 -n.sub.avg)+(1-p.sub.j)(s.sub.j-1 -s.sub.avg) wherein p j is the a posteriori probability of the frame of speech j being noise, n j-1 and s j-1 are the average noise and speech feature vectors for the utterance, and n avg and s avg are the average noise and speech feature vectors for the database of utterances; and computing a normalized feature vector x j as: x.sub.j =x.sub.j -r(x.sub.j). 6.
Claim 7
Independent Claim
183 words
A method for improving speech recognition through front-end normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the method comprising the steps of: providing a database of known utterances, the database of utterances having an average noise feature vector and an average speech feature vector; receiving a feature vector x j representing a frame of speech j in an utterance to be recognized, the frame of speech having an a posteriori probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; computing a correction vector r(x j ) as: r(x.sub.j)=p.sub.j (n.sub.j-1 -n.sub.avg)+(1-p.sub.j)(s.sub.j-1 -s.sub.avg) wherein p j is the a posteriori probability of the frame of speech j being noise, n j-1 and s j-1 are the average noise and speech feature vectors for the utterance, and n avg and s avg are the average noise and speech feature vectors for the database of utterances; and computing a normalized feature vector x j as: x.sub.j =x.sub.j -r(x.sub.j). 6.
Claim 8
Dependent Claim
21 words
The method of claim 5, wherein the step of receiving a feature vector includes the step of receiving a cepstral vector.
Claim 9
Dependent Claim
35 words
The method of claim 5, wherein the a posteriori probability of the frame of speech being noise and the average noise and speech feature vectors for the utterance are updated for each frame of speech.
Claim 10
Dependent Claim
76 words
The method of claim 5, wherein the a posteriori probability of the frame of speech j being noise p j is computed as: ##EQU13## wherein ξ is an a priori probability of the frame of speech j being noise, N(x j ,n j-1 ,ξn j-1 )) and N(x j ,s j-1 ,ξs.sub.(j-1)) are Gaussian probability density functions for noise and speech, respectively, and ξn j-1 ) and ξs.sub.(j-1) are covariance matrices for noise and speech, respectively.
Claim 11
Dependent Claim
50 words
The method of claim 8, wherein the Gaussian probability density functions for noise and speech, N(x j ,n j-1 ,ξn.sub.(j-1)) and N(x j ,s j-1 ,ξs.sub.(j-1), are computed as: ##EQU14## wherein q is a dimension of x j , exp is an exponential function, and T represents a transpose function.
Claim 12
Dependent Claim
38 words
The method of claim 5, wherein the a posteriori probability of the frame of speech j being noise p j is computed as: ##EQU15## wherein d(x j ) is a distortion measure of the frame of speech j.
Claim 13
Dependent Claim
16 words
The method of claim 10, wherein the distortion measure d(x j ) is computed as: ##EQU16##
Claim 14
Dependent Claim
28 words
The method of claim 10, wherein the distortion measure d(x j ) is computed as: ##EQU17## wherein q is a dimension of σ n and σ s .
Claim 15
Dependent Claim
16 words
The method of claim 10, wherein the distortion measure d(x j ) is computed as: ##EQU18##
Claim 16
Dependent Claim
43 words
The method of claim 13, wherein the average noise and speech feature vectors for the utterance are computed as: ##EQU19## wherein w k is an exponential window function represented as: w.sub.k =α.sup.k wherein α is a parameter that controls a rate of adaptation.
Claim 17
Dependent Claim
18 words
The method of claim 14, wherein the diagonal covariance matrices for noise and speech are computed as: ##EQU20##
Claim 18
Dependent Claim
24 words
The method of claim 15, wherein the a priori probability of the frame of speech j being noise ξ j is computed as: ##EQU21##
Claim 19
Dependent Claim
39 words
The method of claim 13, wherein the average noise and speech feature vectors for the utterance are computed as: ##EQU22## and wherein α n and α s are parameters that control rates of adaptation for noise and speech, respectively.
Claim 20
Dependent Claim
18 words
The method of claim 17, wherein the diagonal covariance matrices for noise and speech are computed as: ##EQU23##
Claim 21
Dependent Claim
26 words
The method of claim 18, wherein the a priori probability of the frame of speech j being noise ξ j is computed as: ξ.sub.j =(1-α.sub.n)c.sub.n(j). 20.
Claim 22
Independent Claim
151 words
A system for improving speech recognition through front-send normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the system comprising: a database of known utterances, the database of utterances having an average noise feature vector and an average speech feature vector; and an input normalizer for: receiving a feature vector representing a frame of speech in an utterance to be recognized, the frame of speech having a probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; computing a correction vector based on the probability of the frame of speech being noise and based on the average noise and speech feature vectors for the utterance and the database of utterances; and computing a normalized feature vector based on the feature vector and the correction vector.
Claim 23
Independent Claim
151 words
A system for improving speech recognition through front-send normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the system comprising: a database of known utterances, the database of utterances having an average noise feature vector and an average speech feature vector; and an input normalizer for: receiving a feature vector representing a frame of speech in an utterance to be recognized, the frame of speech having a probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; computing a correction vector based on the probability of the frame of speech being noise and based on the average noise and speech feature vectors for the utterance and the database of utterances; and computing a normalized feature vector based on the feature vector and the correction vector.
Claim 24
Independent Claim
185 words
A system for improving speech recognition through front-end normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the system comprising: a database of known utterances, the utterances being represented by feature models, the database of utterances having an average noise feature vector and an average speech feature vector; a feature extractor for extracting a feature vector from a frame of speech in an utterance to be recognized, the frame of speech having a probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; an input normalizer for normalizing the feature vector by: (i) computing a correction vector based on the probability of the frame of speech being noise and based on the average noise and speech feature vectors for the utterance and the database of utterances, and (ii) computing a normalized feature vector based on the feature vector and the correction vector; and a pattern matcher for comparing the normalized feature vector to the feature models in the database.
Claim 25
Independent Claim
185 words
A system for improving speech recognition through front-end normalization of feature vectors, wherein the speech being recognized comprises utterances, each utterance comprising frames of speech, each frame of speech being represented by a feature vector, the system comprising: a database of known utterances, the utterances being represented by feature models, the database of utterances having an average noise feature vector and an average speech feature vector; a feature extractor for extracting a feature vector from a frame of speech in an utterance to be recognized, the frame of speech having a probability of being noise, the utterance having an average noise feature vector and an average speech feature vector; an input normalizer for normalizing the feature vector by: (i) computing a correction vector based on the probability of the frame of speech being noise and based on the average noise and speech feature vectors for the utterance and the database of utterances, and (ii) computing a normalized feature vector based on the feature vector and the correction vector; and a pattern matcher for comparing the normalized feature vector to the feature models in the database.
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) | 2000-02-18 | 2000-08-19 | 2001-02-18 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2004-02-18 | 2004-08-19 | 2005-02-18 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2008-02-18 | 2008-08-19 | 2009-02-18 | $7700.00 | Window Closed / Lapsed |