US5594834A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method and system for recognizing a boundary between sounds in continuous speech
Assignee: Google Technology Holdings LLC
•
Filing: 1994-09-30
•
Granted: 1997-01-14
•
Expires: 2014-09-30
Abstract
Abstract Boundaries of spoken sounds in continuous speech are identified by classifying delimitative sounds to provide improved performance in a speech-recognition system. Delimitative sounds, those portions of continuous speech that occur between two spoken sounds, are recognized by the same method used to recognize spoken sounds. Recognition of delimitative sounds is accomplished by training a learning machine to act as a classifier which implements a discriminant function based on a polynomial expansion.
Inventors
- 👤 Shay-Ping T. Wang
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-30 03:57
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/04; G10L15/05; G10L15/02; G10L15/06; G10L25/00; G10L25/03; G10L25/18; G10L25/24; G10L25/27; G10L25/30
Claims Analysis
Total Claims: 17
Claim 1
Independent Claim
151 words
In a speech recognition system, a method for recognizing a boundary between two contiguous spoken sounds, the method comprising the steps of: (a) training a learning machine, with a plurality of spoken examples, to classifiy a delimitative sound that represents the boundary; (b) receiving the continuously spoken speech which includes the delimitative sound; (c) extracting a plurality of features from the continuously spoken speech; and (d) recognizing the delimitative sound in the continuously spoken speech by classifying the plurality of features according to a polynomial discriminant function having a form ##EQU3## wherein y represents a classifier output, w i-1 represents a coefficient of an ith term in the polynomial expansion, x 1 , x 2 . . . x n represent the plurality of features g 1i , . . . , g ni represent exponents for the ith term in the polynomial expansion, and i, m, and n are integers.
Claim 2
Dependent Claim
151 words
In a speech recognition system, a method for recognizing a boundary between two contiguous spoken sounds, the method comprising the steps of: (a) training a learning machine, with a plurality of spoken examples, to classifiy a delimitative sound that represents the boundary; (b) receiving the continuously spoken speech which includes the delimitative sound; (c) extracting a plurality of features from the continuously spoken speech; and (d) recognizing the delimitative sound in the continuously spoken speech by classifying the plurality of features according to a polynomial discriminant function having a form ##EQU3## wherein y represents a classifier output, w i-1 represents a coefficient of an ith term in the polynomial expansion, x 1 , x 2 . . . x n represent the plurality of features g 1i , . . . , g ni represent exponents for the ith term in the polynomial expansion, and i, m, and n are integers.
Claim 3
Dependent Claim
13 words
The method of claim 1 wherein the learning machine includes a neural network.
Claim 4
Dependent Claim
138 words
The method of claim 1 wherein training the learning machine further includes a step of calculating a plurality of polynomial coefficients, the step of calculating comprises the following substeps: (i) providing the plurality of spoken examples; (ii) comparing the number of spoken examples in the plurality of spoken examples with the number of the terms in the at least one term of the polynomial expansion; and (iii) providing at least two techniques for calculating the plurality of polynomial coefficients, wherein (A) if the number of the terms is equal to the number of spoken examples, using a matrix-inversion technique to solve for each of the plurality of polynomial coefficients; and (B) if the number of terms is less than the number of spoken examples, using a least-squares estimation technique to solve for each the plurality of polynomial coefficients.
Claim 5
Dependent Claim
24 words
The method of claim 1 wherein the plurality of features corresponds to the continuously spoken speech during an interval of time having a duration.
Claim 6
Dependent Claim
16 words
The method of claim 4 wherein the duration of the interval of time is 10 milliseconds.
Claim 7
Dependent Claim
24 words
The method of claim 1 wherein the plurality of features is selected from the group consisting of cepstral coefficients, predictive coefficients, and Fourier coefficients.
Claim 8
Dependent Claim
127 words
A speech-recognition system comprising: a converter for generating a speech signal representing continuously spoken speech; a feature extractor for extracting a plurality of features from the speech signal; and a classifier for identifying a delimitative sound representing a boundary between two contiguous sounds in the continuously spoken speech, the classifier classifying the plurality of features according to a polynomial discriminant function having a form ##EQU4## wherein y represents a classifier output, w i-1 represents a coefficient of an ith term in the polynomial expansion, x 1 , x 2 , . . . , x n represent the plurality of features, g 1i , . . . , g ni represent exponents for the ith term in the polynomial expansion, and i, m, and n are integers.
Claim 9
Dependent Claim
127 words
A speech-recognition system comprising: a converter for generating a speech signal representing continuously spoken speech; a feature extractor for extracting a plurality of features from the speech signal; and a classifier for identifying a delimitative sound representing a boundary between two contiguous sounds in the continuously spoken speech, the classifier classifying the plurality of features according to a polynomial discriminant function having a form ##EQU4## wherein y represents a classifier output, w i-1 represents a coefficient of an ith term in the polynomial expansion, x 1 , x 2 , . . . , x n represent the plurality of features, g 1i , . . . , g ni represent exponents for the ith term in the polynomial expansion, and i, m, and n are integers.
Claim 10
Dependent Claim
45 words
The speech-recognition system recited in claim 7 further comprising: a learning machine for providing the classifier; a trainer which is operatively coupled to the learning machine, for training the learning machine in order to provide the classifier, the trainer receiving a plurality of spoken examples.
Claim 11
Dependent Claim
108 words
The speech-recognition system of claim 8 wherein the trainer calculates coefficients of a polynomial expansion having at least one term, the trainer comprising: comparing means for comparing the number of spoken examples with the number of terms in the polynomial expansion; and providing means, operatively coupled to the comparing means, for providing at least two techniques for calculating the coefficients, wherein (i) if the number of terms is equal to the number of spoken examples, providing a matrix-inversion technique to solve for the coefficients; (ii) if the number of terms is less than the number of spoken examples, providing a least-squares estimation technique to solve for the coefficients.
Claim 12
Dependent Claim
14 words
The speech-recognition system of claim 7 wherein the recognition means includes a neural network.
Claim 13
Dependent Claim
25 words
The speech-recognition system of claim 7 wherein the plurality of features corresponds to the continuously spoken speech during an interval of time having a duration.
Claim 14
Dependent Claim
17 words
The speech-recognition system of claim 11 wherein the duration of the interval of time is 10 milliseconds.
Claim 15
Dependent Claim
25 words
The speech-recognition system of claim 7 wherein the plurality of features is selected from the group consisting of cepstral coefficients, predictive coefficients, and Fourier coefficients.
Claim 16
Dependent Claim
109 words
The speech-recognition system of claim 7 for identifying a plurality of spoken sounds from the continuously spoken speech, further comprising: a plurality of classifiers for classifying the plurality of spoken sounds, each of the plurality of classifiers having a corresponding class and implementing a discriminant function which uses the plurality of features to produce a result which indicates the likelihood of one of the plurality of spoken sounds belonging to the corresponding class; and a selector, operatively coupled to each of the plurality of classifiers, for identifying of a spoken sound included in the plurality of spoken sounds by comparing the results from each of the plurality of classifiers.
Claim 17
Dependent Claim
19 words
The speech-recognition system of claim 14 wherein at least one of the plurality of classifiers classifies the delimitative sound.
References Cited (Backward Citations)
US4761815A
Cited
US4624008A
Cited
US4718093A
Cited
US4852172A
Cited
US4852180A
Cited
US5072452A
Cited
US5073867A
Cited
US5369726A
Cited
US5404422A
Cited
US5299284A
Cited
US5365592A
Cited
US5247584A
Cited
US5239594A
Cited
US5450522A
Cited
US5305422A
Cited
US5329608A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2000-01-14 | 2000-07-15 | 2001-01-14 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2004-01-14 | 2004-07-15 | 2005-01-14 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2008-01-14 | 2008-07-15 | 2009-01-14 | $7700.00 | Window Closed / Lapsed |