US5506933A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speech recognition using continuous density hidden markov models and the orthogonalizing karhunen-loeve transformation
Assignee: Toshiba Corp
•
Filing: 1993-03-12
•
Granted: 1996-04-09
•
Expires: 2013-03-12
Abstract
Abstract A recognition system comprises a feature extractor for extracting a feature vector x from an input speech signal, and a recognizing section for defining continuous density Hidden Markov Models of predetermined categories k as transition network models each having parameters of transition probabilities p(k,i,j) that a state Si transits to a next state Sj and output probabilities g(k,s) that a feature vector x is output in transition from the state Si to one of the states Si and Sj, and recognizing the input signal on the basis of similarity between a sequence X of feature vectors extracted by the feature extractor and the continuous density HMMs. Particularly, the recognizing section includes a memory section for storing a set of orthogonal vectors φ m (k,s) provided for the continuous density HMMs, and a modified CDHMM processor for obtaining each of the output probabilities g(k,s) for the continuous density HMMs in accordance with corresponding orthogonal vectors φ m (k,s).
Inventors
- 👤 Tsuneo Nitta
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-02 11:17
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/14; G10L15/142; G10L15/144
Claims Analysis
Total Claims: 9
Claim 1
Independent Claim
166 words
A recognition system comprising: feature extracting means for extracting a feature vector x from an input signal; and recognizing means for obtaining continuous density Hidden Markov Models (HMMs) of predetermined categories K represented by transition network models each having parameters of transition probabilities p(k,i,j) that a state Si transits to a next state Sj and output probabilities g(k,s) that the feature vector x is output in transition from the state Si to one of the states Si and Sj, and for recognizing the input signal on the basis of similarity between a feature vector sequence x of the feature vectors x each extracted by said feature extracting means and the continuous density HMMs; wherein said recognizing means includes memory means for storing a set of orthogonal vectors φ m (k,s) provided for the continuous density HMMs, and processing means for obtaining each of the output probabilities g(k,s) for the continuous density HMMs in accordance with the orthogonal vectors φ m (k,s) or a corresponding category k.
Claim 2
Dependent Claim
166 words
A recognition system comprising: feature extracting means for extracting a feature vector x from an input signal; and recognizing means for obtaining continuous density Hidden Markov Models (HMMs) of predetermined categories K represented by transition network models each having parameters of transition probabilities p(k,i,j) that a state Si transits to a next state Sj and output probabilities g(k,s) that the feature vector x is output in transition from the state Si to one of the states Si and Sj, and for recognizing the input signal on the basis of similarity between a feature vector sequence x of the feature vectors x each extracted by said feature extracting means and the continuous density HMMs; wherein said recognizing means includes memory means for storing a set of orthogonal vectors φ m (k,s) provided for the continuous density HMMs, and processing means for obtaining each of the output probabilities g(k,s) for the continuous density HMMs in accordance with the orthogonal vectors φ m (k,s) or a corresponding category k.
Claim 3
Dependent Claim
61 words
A recognition system according to claim 1, wherein said memory means includes a memory section for storing probabilities P(k) that a category k appears, the transition probabilities p(k,i,j), average vectors μ(k,s), the orthogonal vectors φ m (k,s) formed of predetermined eigen vectors contained in covariance matrices C(k,s), and eigen values λ m (k,s), for each continuous density Hidden Markov Model (HMM).
Claim 4
Dependent Claim
115 words
A recognition system according to claim 2, wherein said processing means includes a processing section for obtaining the output probabilities g(k,s) by setting the orthogonal vectors φ m (k,s) into a following equation EQ1, and accumulating a sum of each value of g(k,s) and a corresponding logarithmic value of p(k,i,j) for the feature vector sequence X to produce an accumulated value and an accumulation count for each model, and a discriminator for estimating similarity between the feature vector sequence X and each continuous density HMM from the accumulated value and accumulation count for each model, wherein ##EQU4## where, [·] represents an inner product, and L represents a number of axis of the orthogonal vector set.
Claim 5
Dependent Claim
91 words
A recognition system according to claim 3, further comprising training means for obtaining transition probabilities p(k,i,j), average vectors μ (k,s) and covariance matrices C(k,s) of the feature vector sequence X from said feature extracting means, obtaining upper L eigen vectors serving as the orthogonal vectors φ m (k,s), and eigen values λ m (k,s) by a Karhunen-Loeve (K-L) transformation of the covariance matrices C(k,s), and presetting the orthogonal vectors φ m (k,s), the eigen values λ m (k,s), the average vectors μ(k,s), and the transition probabilities p(k,i,j) in said memory section.
Claim 6
Dependent Claim
49 words
A recognition system according to claim 1, wherein said memory means includes a memory section for storing the transition probabilities p(k,i,j) and the orthogonal vectors φ m (k,s) formed of predetermined eigen vectors contained in correlation matrices R(k,s) of the feature vector sequence X, for each continuous density HMM.
Claim 7
Dependent Claim
114 words
A recognition system according to claim 5, wherein said processing means includes a processing section for obtaining the output probabilities g(k,s) by setting the orthogonal vectors φ m (k,s) into a following equation EQ2, and accumulating a sum of each value of g(k,s) and a corresponding logarithmic value of p(k,i,j) for the feature vector sequence X to produce the accumulated value and the accumulation count for each model, and a discriminator for estimating similarity between the feature vector sequence X and each continuous density HMM from the accumulated value and the accumulating count for each model, ##EQU5## where, represents an inner product, and L represents a number of axis of the orthogonal vector set.
Claim 8
Dependent Claim
72 words
A recognition system according to claim 6, further comprising training means for obtaining transition probabilities p(k,i,j) and the correlation matrices R(k,s) of the feature vector sequence X from said feature extracting means, obtaining upper L eigen vectors serving as the orthogonal vectors φ m (k,s) by a Karhunen-Loeve (K-L) transformation of the correlation matrices R(k,s), and presetting the orthogonal vectors φ m (K,s) and the transition probabilities p(k,i,j) in said memory section.
Claim 9
Dependent Claim
15 words
A recognition system according to claim 1, wherein said input signal is a speech signal.
References Cited (Backward Citations)
US4624011A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1999-04-09 | 1999-10-09 | 2000-04-09 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2003-04-09 | 2003-10-09 | 2004-04-09 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2007-04-09 | 2007-10-09 | 2008-04-09 | $7700.00 | Window Closed / Lapsed |