US5144672A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speech recognition apparatus including speaker-independent dictionary and speaker-dependent
Assignee: Ricoh Co Ltd
•
Filing: 1990-09-28
•
Granted: 1992-09-01
•
Expires: 2010-09-28
Abstract
Abstract A speech recognition apparatus includes an input unit, a pattern generator, a speaker-independent dictionary for storing a first group of reference patterns, and a speaker-dependent dictionary for storing a second group of reference patterns. The apparatus also includes a dictionary information storing unit which stores dictionary information about the first group of reference patterns and the second group of reference patterns, and a weighting unit which calculates a weighting factor from the dictionary information and which weights the second group of reference patterns in the speaker-dependent dictionary by using the weighting factor to thereby output weighted reference patterns. Further, the apparatus includes a recognition unit which recognizes the speech pattern generated by the pattern generator by using the first group of reference patterns in the speaker-independent dictionary and the weighted reference patterns supplied from the weighting unit, so that the first group of reference patterns and the weighted reference patterns are equally handled.
Inventors
- 👤 Shoji Kuriki
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-29 13:16
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/06; G10L15/063
Claims Analysis
Total Claims: 9
Claim 1
Independent Claim
292 words
A speech recognition apparatus comprising: input means for receiving an input speech signal and for extracting features from said input speech; pattern generating means, coupled to said input means, for generating a speech pattern of said input speech from said features extracted by said input means, said speech pattern having a time-spectrum pattern showing frequency elements obtained at predetermined intervals; a speaker-independent dictionary for storing a first group of reference patterns, each obtained by adding a plurality of time-spectrum patterns of an identical word uttered by different speakers; a speaker-dependent dictionary for storing a second group of reference patterns, each obtained by adding time-spectrum patterns of an identical word uttered by an identical speaker; dictionary information storing means, coupled to said speaker-independent dictionary and said speaker-dependent dictionary, for storing dictionary information about said first group of reference patterns and said second group of reference patterns, the dictionary information storing means coupled to said pattern generating means so as to receive the speech pattern generated by the pattern generating means; weighting means, coupled to said speaker-dependent dictionary and said dictionary information storing means, for calculating a weighting factor from said dictionary information and for weighting said second group of reference patterns in said speaker-dependent dictionary by using said weighting factor to thereby output weighted reference patterns; and recognizing means, coupled to said pattern generating means, said speaker-independent dictionary and said weighting means, for recognizing said speech pattern of said input speech signal generated by said pattern generating means by using said first group of reference patterns in said speaker-independent dictionary and said weighted reference patterns supplied from said weighting means, so that said first group of reference patterns and said weighted reference patterns are equally handled, said recognizing means outputting a recognition result.
Claim 2
Dependent Claim
292 words
A speech recognition apparatus comprising: input means for receiving an input speech signal and for extracting features from said input speech; pattern generating means, coupled to said input means, for generating a speech pattern of said input speech from said features extracted by said input means, said speech pattern having a time-spectrum pattern showing frequency elements obtained at predetermined intervals; a speaker-independent dictionary for storing a first group of reference patterns, each obtained by adding a plurality of time-spectrum patterns of an identical word uttered by different speakers; a speaker-dependent dictionary for storing a second group of reference patterns, each obtained by adding time-spectrum patterns of an identical word uttered by an identical speaker; dictionary information storing means, coupled to said speaker-independent dictionary and said speaker-dependent dictionary, for storing dictionary information about said first group of reference patterns and said second group of reference patterns, the dictionary information storing means coupled to said pattern generating means so as to receive the speech pattern generated by the pattern generating means; weighting means, coupled to said speaker-dependent dictionary and said dictionary information storing means, for calculating a weighting factor from said dictionary information and for weighting said second group of reference patterns in said speaker-dependent dictionary by using said weighting factor to thereby output weighted reference patterns; and recognizing means, coupled to said pattern generating means, said speaker-independent dictionary and said weighting means, for recognizing said speech pattern of said input speech signal generated by said pattern generating means by using said first group of reference patterns in said speaker-independent dictionary and said weighted reference patterns supplied from said weighting means, so that said first group of reference patterns and said weighted reference patterns are equally handled, said recognizing means outputting a recognition result.
Claim 3
Dependent Claim
55 words
A speech recognition apparatus as claimed in claim 1, wherein said dictionary information comprises: a first maximum value of said frequency elements in said first group of reference patterns registered in said speaker-independent dictionary; and a second maximum value of said frequency elements in said second group of reference patterns registered in said speaker-dependent dictionary.
Claim 4
Dependent Claim
95 words
A speech recognition apparatus as claimed in claim 2, wherein said weighting means comprises: dividing means for dividing said first maximum value by said second maximum value and for outputting a divided result; integer extracting means, coupled to said dividing means, for extracting an integer from said divided result; and multiplying means, coupled to said speaker-dependent dictionary and said integer extracting means, for multiplying one of said second group of reference patterns registered in said speaker-dependent dictionary by said integer and for outputting a multiplied result which corresponds to one of said weighted reference patterns.
Claim 5
Dependent Claim
152 words
A speech recognition apparatus as claimed in claim 2, wherein said weighting means comprises: table memory means for storing predetermined weighting factors defined by possible first maximum values of said frequency elements of said first group of reference patterns and possible second maximum values of said frequency elements of said second group of reference patterns and for outputting one of said predetermined weighting factors defined by one of said possible first maximum values corresponding to said first maximum value and one of said possible second maximum values corresponding to said second maximum value, said one of the predetermined weighting factors corresponding to said weighting factor; and multiplying means, coupled to said speaker-dependent dictionary and said table memory means, for multiplying one of said second group of reference patterns registered in said speaker-dependent dictionary by said weighting factor and for outputting a multiplied result which corresponds to one of said weighted reference patterns.
Claim 6
Dependent Claim
63 words
A speech recognition apparatus as claimed in claim 1, wherein said dictionary information comprises: a first number corresponding to said plurality of patterns of said identical word used for creating one of said first group of reference patterns; and a second number corresponding to said plurality of patterns of said identical word used for creating one of said second group of reference patterns.
Claim 7
Dependent Claim
93 words
A speech recognition apparatus as claimed in claim 5, wherein: said weighting means comprises: dividing means for dividing said first number by said second number and for outputting a divided result; integer extracting means, coupled to said dividing means, for extracting an integer from said divided result; and multiplying means, coupled to said speaker-dependent dictionary and said integer extracting means, for multiplying one of said second group of reference patterns registered in said speaker-dependent dictionary by said integer and for outputting a multiplied result which corresponds to one of said weighted reference patterns.
Claim 8
Dependent Claim
152 words
A speech recognition apparatus as claimed in claim 5, wherein said weighting means comprises: table memory means for storing predetermined weighting factors defined by possible first maximum values of said frequency elements of said first group of reference patterns and possible second maximum values of said frequency elements of said second group of reference patterns and for outputting one of said predetermined weighting factors defined by one of said possible first maximum values corresponding to said first maximum value and one of said possible second maximum values corresponding to said second maximum value, said one of the predetermined weighting factors corresponding to said weighting factor; and multiplying means, coupled to said speaker-dependent dictionary and said table memory means, for multiplying one of said second group of reference patterns registered in said speaker-dependent dictionary by said weighting factor and for outputting a multiplied result which corresponds to one of said weighted reference patterns.
Claim 9
Dependent Claim
41 words
A speech recognition apparatus as claimed in claim 1, wherein said dictionary information storing means stores said dictionary information at the same time as said first and second groups of reference patterns are stored in said speaker-independent and speaker-dependent dictionaries, respectively.
References Cited (Backward Citations)
DE3216800A1
Cited
DE3129282A1
Cited
DE3129353A1
Cited
US4694493A
Cited
EP0191354A1
Cited
US4910782A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1995-09-01 | 1996-03-02 | 1996-09-01 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1999-09-01 | 2000-03-02 | 2000-09-01 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2003-09-01 | 2004-03-02 | 2004-09-01 | $7700.00 | Window Closed / Lapsed |