EP0241163A1
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Speaker-trained speech recognizer
Assignee: AT&T Inc
•
Filing: 1987-03-18
•
Granted: —
•
Expires: 2007-03-18
Abstract
Abstract During a training sequence, a speaker-trained speech recognizer detects and signals the speaker when vocabulary word pairs are p'otentially confusing to the recognizer. Each vocabulary word is converted (106) into parameters representing a predetermined reference model of that word. The feature signals of a subsequent potential vocabulary word are compared against the reference model of each vocabulary word previously stored in the recognizer memory (105). The speaker is signaled (107) when the potential vocabulary word is confusingly similar to one of the existing vocabulary words.
Inventors
- 👤 Bruce Allen Dautrich
- 👤 Thomas William Goeddel
- 👤 David Bjorn Roe
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-01 12:31
CPC Classifications
G; G10; G10L; G10L13/00; G10L15/00; G10L15/06; G10L15/065; G10L15/07
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
113 words
A speaker-trained speech recognizer characterized by means for extracting (106) a plurality of feature signals from a present word utterance received from a speaker, means for generating (101, 206) a plurality of parameters from said plurality of feature signals, means for comparing (101,211) said plurality of feature signals of said present word utterance derived from said extracting means against a plurality of parameters of a previous word reference model derived from a previously received word utterance using a predetermined criterion, and means for signaling (107) said speaker when an output of said comparing means indicates that the difference between said present word utterance and the reference model is less than a predetermined value.
Claim 2
Dependent Claim
113 words
A speaker-trained speech recognizer characterized by means for extracting (106) a plurality of feature signals from a present word utterance received from a speaker, means for generating (101, 206) a plurality of parameters from said plurality of feature signals, means for comparing (101,211) said plurality of feature signals of said present word utterance derived from said extracting means against a plurality of parameters of a previous word reference model derived from a previously received word utterance using a predetermined criterion, and means for signaling (107) said speaker when an output of said comparing means indicates that the difference between said present word utterance and the reference model is less than a predetermined value.
Claim 3
Dependent Claim
17 words
The speaker-trained speech recognizer of claim 1 wherein said predetermined reference model is a Hidden Markov Model.
Claim 4
Dependent Claim
19 words
The speaker-trained speech recognizer of claim 1 or claim 2 wherein said extracting means comprises a filter bank means.
Claim 5
Dependent Claim
20 words
4, The speaker-trained speech recognizer of any of the preceding claims wherein said comparing means comprises a Viterbi scoring means.
Claim 6
Dependent Claim
96 words
The speaker-trained speech recognizer of any of the preceding claims wherein said comparing means comprises a previous word model scoring means (304) wherein said plurality of feature signals of said present word utterance are scored against each plurality of parameters of a previous reference word model derived from each previously received word utterance using said predetermined criterion, and said comparing means generates a similarity index (305) using the lowest score from said previous word model scoring means, and wherein said signaling means signals said speaker when said similarity index is less than a predetermined value (311).
Claim 7
Dependent Claim
69 words
The speaker-trained speech recognizer of claim 5 wherein said comparing means includes present word model scoring means wherein said plurality of feature signals of said present word utterance is scored against said predetermined reference model derived therefrom (308,309) and wherein said comparing means utilizes an output of said present word model scoring means and the outputs of said previous word model scoring means to determine said similarity index (310).
Claim 8
Dependent Claim
40 words
The speaker-trained speech recognizer of claim 6 wherein said lowest score from said previous word scoring means is S(MIN), said output of said present word model scoring means is S(N+1), and said similarity index is equal to S(N+1) - S(MIN).
Claim 9
Dependent Claim
40 words
The speaker-trained speech recognizer of claim 6 wherein said lowest score from said previous word scoring means is S(MIN), said output of said present word scoring means is S(N+1), and said similarity index is equal to S(N+1) divided by S(MIN).
Claim 10
Dependent Claim
31 words
The speaker-trained speech recognizer of claim 5 wherein said lowest score from said previous word scoring means is S(MIN), and said similarity index is equal to a duration-normalized form of S(MIN).
Claim 11
Dependent Claim
33 words
The speaker-trained speech recognizer of any of the preceding claims including means for updating said previous reference word model in response to said plurality of feature signals of said present word utterance (314).
Claim 12
Independent Claim
108 words
A method of operating a speaker-trained speech recognizer characterized by extracting (106) a plurality of feature signals from a present word utterance received from a speaker, generating (101,206) a plurality of parameters from said plurality of feature signals, comparing (101,211) said plurality of feature signals of said present word utterance derived from said extracting means against a plurality of parameters of a previous word reference model derived from a previously received word utterance using a predetermined criterion, and signaling (107) said speaker when an output of said comparing means indicates that the difference between said present word utterance and the reference model is less than a predetermined value.
Claim 13
Independent Claim
108 words
A method of operating a speaker-trained speech recognizer characterized by extracting (106) a plurality of feature signals from a present word utterance received from a speaker, generating (101,206) a plurality of parameters from said plurality of feature signals, comparing (101,211) said plurality of feature signals of said present word utterance derived from said extracting means against a plurality of parameters of a previous word reference model derived from a previously received word utterance using a predetermined criterion, and signaling (107) said speaker when an output of said comparing means indicates that the difference between said present word utterance and the reference model is less than a predetermined value.
References Cited (Backward Citations)
US3548202A
Cited
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.