US4852172A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speech recognition system
Assignee: NEC Corp
•
Filing: 1986-09-02
•
Granted: 1989-07-25
•
Expires: 2006-09-02
Abstract
Abstract A speech recognition system for recognizing the words spoken by identified speakers and unidentified speakers, based upon the use of three dimensional, polar coordinate expressions of formant information, which eliminate the differences in word patterns encountered among various speakers, and the content of a reference pattern memory which stores formant information as a reference pattern in polar coordinates. The input speech is subjected to LPC analysis for each of several analysis degrees and the resulting coefficients are time normalized against the reference pattern using a DP matching process, which may be on the basis of city block distances in order to determine the most appropriate analysis degree that represents the input speech.
Inventors
- 👤 Tetsu Taguchi
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-29 14:18
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/12; G10L15/06; G10L15/065; G10L15/07
Claims Analysis
Total Claims: 8
Claim 1
Independent Claim
205 words
A speech recognition system for recognizing the content of an input speech signal comprising: a reference pattern memory for storing reference patterns each being time-serial feature vectors of groups comprising at least first through third formant frequency information for a plurality of analysis frames; first means for developing LPC coefficients for each one of a predetermined plurality of analysis degrees, in each of said plurality of analysis frames, from an input pattern of said input speech signal; second means for developing first through third formant frequencies, for each of said plurality of analysis degrees in each of said plurality of analysis frames, from the developed LFC coefficients; third means for determining distances between said reference patterns and said input pattern, said input pattern being comprised of time-serially connected feature vectors for a plurality of groups of first through third formant frequency information for said plurality of frames, and each of said, groups being one of the groups of first through third formant frequency information developed for said plurality of analysis degrees in each frame; and fourth means for determining the reference pattern having the shortest distance from said input pattern on the basis of the distance determined by said third means as said input pattern.
Claim 2
Dependent Claim
205 words
A speech recognition system for recognizing the content of an input speech signal comprising: a reference pattern memory for storing reference patterns each being time-serial feature vectors of groups comprising at least first through third formant frequency information for a plurality of analysis frames; first means for developing LPC coefficients for each one of a predetermined plurality of analysis degrees, in each of said plurality of analysis frames, from an input pattern of said input speech signal; second means for developing first through third formant frequencies, for each of said plurality of analysis degrees in each of said plurality of analysis frames, from the developed LFC coefficients; third means for determining distances between said reference patterns and said input pattern, said input pattern being comprised of time-serially connected feature vectors for a plurality of groups of first through third formant frequency information for said plurality of frames, and each of said, groups being one of the groups of first through third formant frequency information developed for said plurality of analysis degrees in each frame; and fourth means for determining the reference pattern having the shortest distance from said input pattern on the basis of the distance determined by said third means as said input pattern.
Claim 3
Dependent Claim
18 words
A speech recognition system according to claim 1, wherein said predetermined plurality of analysis degrees are continuous numbers.
Claim 4
Dependent Claim
18 words
A speech recognition system according to claim 1, wherein said predetermined plurality of analysis degrees are interrupted numbers.
Claim 5
Dependent Claim
20 words
A speech recognition system according to claim 1, wherein said plurality of analysis frames is 8 to 14 in number.
Claim 6
Dependent Claim
16 words
A speech recognition system according to claim 1, wherein said plurality of analysis frames is variable.
Claim 7
Dependent Claim
209 words
A formant information extractor of an input speech signal comprising: a reference pattern memory for storing reference patterns each being time-serial feature vectors of groups comprising at least first through third formant frequency information for a plurality of analysis frames: first means for developing LPC coefficients for each one of a predetermined plurality of analysis degree, in each of said plurality of analysis frames, from an input pattern of said input speech signal; second means for developing first through third formant frequencies, for each of a said plurality of analysis degrees in each of said plurality of analysis frames, from the developed LPC coefficients; third means for determining distances between said reference patterns and said input pattern, said input pattern being comprised of time-serially connected feature vectors for a plurality of groups of first through third formant frequency information for said plurality of frames, and each of said groups being one of the groups of first through third formant frequency information developed for said plurality of analysis degrees in each frame; and fourth means for determining the formant frequency information of the reference pattern having the shortest distance from said input pattern of the basis of the distance determined by said third means as that of said input pattern.
Claim 8
Dependent Claim
209 words
A formant information extractor of an input speech signal comprising: a reference pattern memory for storing reference patterns each being time-serial feature vectors of groups comprising at least first through third formant frequency information for a plurality of analysis frames: first means for developing LPC coefficients for each one of a predetermined plurality of analysis degree, in each of said plurality of analysis frames, from an input pattern of said input speech signal; second means for developing first through third formant frequencies, for each of a said plurality of analysis degrees in each of said plurality of analysis frames, from the developed LPC coefficients; third means for determining distances between said reference patterns and said input pattern, said input pattern being comprised of time-serially connected feature vectors for a plurality of groups of first through third formant frequency information for said plurality of frames, and each of said groups being one of the groups of first through third formant frequency information developed for said plurality of analysis degrees in each frame; and fourth means for determining the formant frequency information of the reference pattern having the shortest distance from said input pattern of the basis of the distance determined by said third means as that of said input pattern.
References Cited (Backward Citations)
US3989896A
Cited
US4516215A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1992-07-24 | 1993-01-23 | 1993-07-25 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1996-07-24 | 1997-01-23 | 1997-07-25 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2000-07-24 | 2001-01-23 | 2001-07-25 | $7700.00 | Window Closed / Lapsed |