US6138095A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speech recognition
Assignee: Nokia of America Corp
•
Filing: 1998-09-03
•
Granted: 2000-10-24
•
Expires: 2018-09-03
Abstract
Abstract Speech recognition in which the log probabilities of the null and alternative hypothesis are computed for an input speech sample by comparison with specific stored speech vocabularies/grammars and with general speech characteristics. The difference in probabilities is normalized by the magnitude of the null hypothesis to derive a likelihood factor which is compared with a rejection threshold that is utterance-length dependent. Advantageously, a high-order polynomial representation of the rejection threshold length dependency may be simplified by a series of piece-wise constants which are stored as rejection thresholds to be selected in accordance with the length of the input speech sample.
Inventors
- 👤 Sunil K. Gupta
- 👤 Frank Kao-Ping Soong
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 22:38
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08
Claims Analysis
Total Claims: 10
Claim 1
Independent Claim
138 words
A method of recognizing speech encoded as electrical signals, comprising: a) processing input speech signals to extract one or more feature vectors; b) comparing the extracted feature vectors with stored speech models and a task grammar to derive a null hypothesis factor indicating the probability that the input speech is correctly recognized; c) comparing the extracted feature vectors with a general speech model and a garbage loop grammar to derive an alternate hypothesis factor indicating the probability that the extracted feature vectors correspond to the characteristics of general speech; d) normalizing the log difference of the null and alternate hypothesis probability factors by the magnitude of the log likelihood of the null hypothesis factor; and e) rejecting the speech where the difference of the normalized probability factors is less than a rejection factor derived from the utterance length.
Claim 2
Independent Claim
138 words
A method of recognizing speech encoded as electrical signals, comprising: a) processing input speech signals to extract one or more feature vectors; b) comparing the extracted feature vectors with stored speech models and a task grammar to derive a null hypothesis factor indicating the probability that the input speech is correctly recognized; c) comparing the extracted feature vectors with a general speech model and a garbage loop grammar to derive an alternate hypothesis factor indicating the probability that the extracted feature vectors correspond to the characteristics of general speech; d) normalizing the log difference of the null and alternate hypothesis probability factors by the magnitude of the log likelihood of the null hypothesis factor; and e) rejecting the speech where the difference of the normalized probability factors is less than a rejection factor derived from the utterance length.
Claim 3
Dependent Claim
43 words
A method of recognizing speech according to claim 1, wherein the rejection factor comprises a threshold modeled as a polynomial in utterance length T, and wherein a separate rejection threshold v i , is derived for each interval representing the input utterance lengths.
Claim 4
Dependent Claim
19 words
A method of recognizing speech according to claim 2, wherein said polynomial is approximated by a piecewise constant approximation.
Claim 5
Dependent Claim
154 words
A speech recognition system comprising: a) a memory for storing: i) a plurality of recognition models each representing a known word, sub-word, phoneme or digit within a vocabulary or task grammar; ii) a general speech model and garbage loop grammar, and iii) a plurality of rejection thresholds modeled as polynomials differing according to the lengths of input utterance and stored as piecewise constant approximations of said polynomials; and b) a processor for receiving an input signal representing a speech utterance and for processing the input signal utilizing the stored models to calculate the respective probabilities that the input signal corresponds to a particular one of the recognition models and to said general speech model; generating a difference signal normalized by the magnitude of one of the probabilities; determining the length of the input utterance; and selecting one of the stored piecewise approximations of said rejection thresholds appropriate to the length of the input utterance.
Claim 6
Dependent Claim
154 words
A speech recognition system comprising: a) a memory for storing: i) a plurality of recognition models each representing a known word, sub-word, phoneme or digit within a vocabulary or task grammar; ii) a general speech model and garbage loop grammar, and iii) a plurality of rejection thresholds modeled as polynomials differing according to the lengths of input utterance and stored as piecewise constant approximations of said polynomials; and b) a processor for receiving an input signal representing a speech utterance and for processing the input signal utilizing the stored models to calculate the respective probabilities that the input signal corresponds to a particular one of the recognition models and to said general speech model; generating a difference signal normalized by the magnitude of one of the probabilities; determining the length of the input utterance; and selecting one of the stored piecewise approximations of said rejection thresholds appropriate to the length of the input utterance.
Claim 7
Dependent Claim
21 words
The system as set forth in claim 4 wherein said plurality of recognition models includes one or more Hidden Markov Models.
Claim 8
Independent Claim
108 words
A method for improving the accuracy of a speech recognition system utilizing a plurality of recognition models to identify whether input speech corresponds to phonemes within a task grammar or to general speech characteristics, comprising the steps of: generating a signal representing the log difference in likelihoods that the input signal corresponds to one of said phonemes or to said general speech characteristics; normalizing said difference signal according to the magnitude of the log likelihood of said general speech characteristics; modeling a rejection threshold as a polynomial in utterance length; and rejecting said input speech when said difference signal does not exceed said utterance length dependant rejection threshold.
Claim 9
Independent Claim
108 words
A method for improving the accuracy of a speech recognition system utilizing a plurality of recognition models to identify whether input speech corresponds to phonemes within a task grammar or to general speech characteristics, comprising the steps of: generating a signal representing the log difference in likelihoods that the input signal corresponds to one of said phonemes or to said general speech characteristics; normalizing said difference signal according to the magnitude of the log likelihood of said general speech characteristics; modeling a rejection threshold as a polynomial in utterance length; and rejecting said input speech when said difference signal does not exceed said utterance length dependant rejection threshold.
Claim 10
Dependent Claim
21 words
The method as set forth in claim 6 wherein said plurality of recognition models includes one or more hidden Markov models.
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) | 2003-10-24 | 2004-04-24 | 2004-10-24 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2007-10-24 | 2008-04-24 | 2008-10-24 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2011-10-24 | 2012-04-24 | 2012-10-24 | $7700.00 | Window Closed / Lapsed |