Patent US5029212A successfully refreshed from Google Patents.
US5029212A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Continuous speech recognition unit using forward probabilities
Assignee: NEC Corp
•
Filing: 1989-10-03
•
Granted: 1991-07-02
•
Expires: 2009-10-03
Abstract
Abstract A continuous speech recognition unit using forward probabilities for recognizing continuous speech associated with standard patterns for given units of recognition comprises a standard template memory for storing Markov model standard templates of standard speech, which are composed of state sequences and transition probabilities between the states; an observation probability computing device for computing a forward probability for a feature vector time sequence; and a cumulative value computing device for determining a cumulative value based on the sum of previous cumulative values. The unit further comprises a matching pass memory for storing maximum values produced by the cumulative value computing means and a result processor for determining recognition results indicative of recognized words. The unit stores the transition giving the best probability in memory for each state and traces back the recognition result for the word sequence based on the transitions in memory.
Inventors
- 👤 Kazunaga Yoshida
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-05 19:00
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/14; G10L15/142
Claims Analysis
Total Claims: 9
Claim 1
Independent Claim
211 words
A continuous speech recognition unit using forward probabilities, comprising: a standard pattern memory for storing a plurality of standard patterns, having a form of Markov model templates represented by an observation probability density function providing a plurality of states and a plurality of transition probabilities between said states, in a predetermined manner; probability computing means for determining an observation probability at each point in time for a feature vector time sequence input to said probability computing means by use of said observation probability density function; cumulative value computing means for computing a new cumulative value based on a product of said observation probability, at least one of said transition probabilities stored in said standard pattern memory and a sum of cumulative values for all transitions for each time of said feature vector time sequence; a matching pass memory for storing, for each of said states, transition information giving the maximum value of said product of said observation probability, said at least one transition probability and said cumulative value for all transitions in said states; and result processor means for determining the most probable ones of said states by back-tracing transition information for each of said states stored in said matching pass memory so as to obtain a plurality of recognition results.
Claim 2
Dependent Claim
211 words
A continuous speech recognition unit using forward probabilities, comprising: a standard pattern memory for storing a plurality of standard patterns, having a form of Markov model templates represented by an observation probability density function providing a plurality of states and a plurality of transition probabilities between said states, in a predetermined manner; probability computing means for determining an observation probability at each point in time for a feature vector time sequence input to said probability computing means by use of said observation probability density function; cumulative value computing means for computing a new cumulative value based on a product of said observation probability, at least one of said transition probabilities stored in said standard pattern memory and a sum of cumulative values for all transitions for each time of said feature vector time sequence; a matching pass memory for storing, for each of said states, transition information giving the maximum value of said product of said observation probability, said at least one transition probability and said cumulative value for all transitions in said states; and result processor means for determining the most probable ones of said states by back-tracing transition information for each of said states stored in said matching pass memory so as to obtain a plurality of recognition results.
Claim 3
Dependent Claim
69 words
Continuous speech recognition unit as claimed in claim 1, wherein said cumulative value computing means computes, as said cumulative value, a product of a sum of said cumulative values of said observation probabilities obtained for a plurality of matching passes wherein said recognition results are estimated by said result processor means are coincident with a plurality of recognition results for said matching passes which give a maximum observation probability.
Claim 4
Dependent Claim
16 words
Continuous speech recognition unit as claimed in claim 2, wherein said observation probability is forward likelihood.
Claim 5
Dependent Claim
16 words
Continuous speech recognition unit as claimed in claim 1, wherein said observation probability is forward likelihood.
Claim 6
Dependent Claim
173 words
The continuous speech recognition unit of claim 1, wherein said cumulative value computing means connects the standard patterns prepared for each of a plurality of recognition units, each of said recognition units representing a word, to one another according to a predetermined network, said cumulative value computing means further comprising: means for determining a plurality of input times corresponding to a plurality of starting points of said standard patterns of said recognition units from said transition information based on a product of said observation probability, said transition probability and said cumulative value determined for transitions of said feature vector up to one of said input times as a new cumulative value; means for determining, from said transitions, one of said transitions producing a maximum value of the products of said observation probability, said transition probability between said states and said cumulative value up to said time at which said sum is obtained, and the sum of said transitions coinciding in input times with one another, said input times corresponding to said starting point.
Claim 7
Dependent Claim
16 words
Continuous speech recognition unit as claimed in claim 5, wherein said observation probability is forward likelihood.
Claim 8
Dependent Claim
150 words
The continuous speech recognition unit of claim 1, wherein said cumulative value computing means connects said standard patterns prepared for each of a plurality of recognition units, such as word, to one another according to a predetermined network, said cumulative value computing means further comprising: means for determining the most likely of said recognition results when said new cumulative value is determined based on a product of said observation probability, said transition probability and previous ones of said cumulative values up to this time in said states for one feature vector; and means for determining one of said transitions producing a maximum value within said transitions of said products of said observation probability and said cumulative value up to said time at which the sum is obtained, in the states and sum of the transitions whose recognition results are at least partially coincident with each other up to this time.
Claim 9
Dependent Claim
16 words
Continuous speech recognition unit as claimed in claim 7, wherein said observation probability is forward likelihood.
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) | 1994-07-01 | 1994-12-31 | 1995-07-02 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1998-07-01 | 1998-12-31 | 1999-07-02 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2002-07-01 | 2002-12-31 | 2003-07-02 | $7700.00 | Window Closed / Lapsed |