Patent EP0242743A1 successfully refreshed from Google Patents.
EP0242743A1
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Speech recognition system
Assignee: Texas Instruments Inc
•
Filing: 1987-04-10
•
Granted: —
•
Expires: 2007-04-10
Abstract
Abstract A speech recognition system (10) uses a word hypothesizer (12) to generate hypothesized words for a sentence recognizer (14). The sentence recognizer (14) receives information containing the hypothesized word and its beginning and end points. The sentence recognizer (14) compares each incoming hypothesis with words allowed by a grammar, and generates a partial or complete sentence hypothesis for each state in which the hypothesis is allowed by the grammar. A dynamic programming technique using backpointers is used to retain all currently valid partial and complete sentence hypotheses. When a complete sentence satisfies suitable criteria, it is output as a recognized sentence. Gaps and overlaps between hypothesized words are used in calculation of a sentence error for each partial and complete sentence hypothesis.
Inventors
- 👤 William M. Fisher
- 👤 Michael L. Mcmahan
- 👤 George R. Doddington
- 👤 Enrico L. Bocchieri
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-10-29 20:47
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/18; G10L15/183; G10L15/19; G10L15/193
Claims Analysis
Total Claims: 33
Claim 1
Independent Claim
35 words
A speech recognition system, comprising: a word recognizer for generating word hypotheses from speech input; and a sentence recognizer coupled to said word hypothesizer for generating partial and complete sentence hypotheses from the word hypotheses.
Claim 2
Dependent Claim
35 words
A speech recognition system, comprising: a word recognizer for generating word hypotheses from speech input; and a sentence recognizer coupled to said word hypothesizer for generating partial and complete sentence hypotheses from the word hypotheses.
Claim 3
Dependent Claim
27 words
The system of Claim 1, wherein said word recognizer generates word hypotheses by comparing the speech input to reference templates corresponding to words in a selected vocabulary.
Claim 4
Dependent Claim
47 words
The system of Claim 2, wherein each word hypothesis contains an identification of the hypothesized word, an error value indicating the closeness of the match between the reference template for the hypothesized word and the speech input, and the beginning and end times of the hypothesized word.
Claim 5
Dependent Claim
21 words
The system of Claim 1, wherein the word hypotheses generated by said word hypothesizer are not constrained by the sentence recognizer.
Claim 6
Dependent Claim
43 words
The system of Claim 1, wherein said sentence recognizer compares each word hypothesis to a grammar defining allowable sentences, and generates partial and complete sentence hypotheses for all partial and complete sentences which validly end with the hypothesized word according to the grammar.
Claim 7
Dependent Claim
41 words
The system of Claim 1, wherein said .sentence recognizer compares each word hypothesis to a grammar defining allowable sentences, and generates a subset of all valid partial and complete sentences hypotheses which end with the hypothesized word according to the grammar.
Claim 8
Dependent Claim
49 words
The system of Claim 1, wherein said sentence recognizer compares each word hypothesis to a grammar defining allowable sentences, and generates one partial or complete sentence hypothesis for the partial or complete sentence which validly ends with the hypothesized word and completes a best hypothesis according to the grammar.
Claim 9
Independent Claim
106 words
A method for recognizing speech, comprising the steps of: (a) receiving input signals representative of speech; (b) comparing the input signals to reference templates representing a vocabulary of allowable words; (c) generating a plurality of word hypotheses based on the comparisons of step (b); (d) comparing the word hypotheses to a grammar specification to determine allowable partial sentences ending in each of the hypothesized words; (e) linking hypothesized words into partial and complete sentences according to the grammar specification; (f) calculating partial and complete sentence errors for all partial and complete sentences; and (g) selecting a complete sentence hypothesis based on the results of step (f).
Claim 10
Independent Claim
106 words
A method for recognizing speech, comprising the steps of: (a) receiving input signals representative of speech; (b) comparing the input signals to reference templates representing a vocabulary of allowable words; (c) generating a plurality of word hypotheses based on the comparisons of step (b); (d) comparing the word hypotheses to a grammar specification to determine allowable partial sentences ending in each of the hypothesized words; (e) linking hypothesized words into partial and complete sentences according to the grammar specification; (f) calculating partial and complete sentence errors for all partial and complete sentences; and (g) selecting a complete sentence hypothesis based on the results of step (f).
Claim 11
Dependent Claim
90 words
The method of Claim 8, wherein step (f) comprises the steps of: (f.1) determining a time difference between the beginning of a current word hypothesis and the end of the previous word hypothesis of the partial sentence hypothesis; (f.2) assigning an error to any difference determined in step (f.1); (f.3) summing the error assigned in step (f.2) with an error representing closeness of fit between the hypothesized word and the actual input speech and with a partial sentence error calculated for the partial sentence ending in the previous word hypothesis.
Claim 12
Dependent Claim
44 words
The method of Claim 9,° further comprising the step of: (f.4) optimizing each partial and complete sentence hypothesis over its entire lenght, wherein partial and complete sentence hypothesis having a lower total error are retained in preference to those having a higher total error.
Claim 13
Dependent Claim
26 words
The method of Claim 8, wherein the vocabulary of reference templates includes templates representing transitional sounds commonly generated between selected word pairs allowed by the grammar.
Claim 14
Independent Claim
104 words
A method for calculating a sentence error for a sentence formed from a sequence of hypothesized words, comprising the steps of; (a) determining beginning and end points for each hypothesized word; (b) determining a word error for each word which indicates a likelihood of the hypothesized word having been incorrectly hypothesized; (c) determining whether any gaps occur between consecutive hypothesized words; (d) assigning error values to any gaps determined in step (c); (e) determining whether any overlap occurs between consecutive hypothesized words; (f) assigning error values to any overlaps determined in step (e); and (g) summing the word errors, gap errors and overlap errors.
Claim 15
Independent Claim
104 words
A method for calculating a sentence error for a sentence formed from a sequence of hypothesized words, comprising the steps of; (a) determining beginning and end points for each hypothesized word; (b) determining a word error for each word which indicates a likelihood of the hypothesized word having been incorrectly hypothesized; (c) determining whether any gaps occur between consecutive hypothesized words; (d) assigning error values to any gaps determined in step (c); (e) determining whether any overlap occurs between consecutive hypothesized words; (f) assigning error values to any overlaps determined in step (e); and (g) summing the word errors, gap errors and overlap errors.
Claim 16
Dependent Claim
92 words
The method of Claim 12, futher comprising the steps of: - (h) determining whether a gap occurs between the beginning of the first hypothesized word and the beginning of the sentence; (i) assigning an error value to any gap determined in step (h); (j) determining whether a gap occurs between the end of the last hypothesized word and the end of the sentence; (k) assigning an error value to any gap determined in step (j); and (I) including the errors assigned in steps (i) and (k) with the summation of step (g).
Claim 17
Dependent Claim
39 words
The method of Claim 13, wherein gap errors and word errors are assigned as each word is hypothesized, and wherein a partial sentence error is calculated by summing all errors so far assigned each time a word is hypothesized.
Claim 18
Dependent Claim
28 words
The method of Claim 12, wherein gap errors are assigned in step (d) only to those gaps which correspond to speech which did not generate a hypothesized word.
Claim 19
Dependent Claim
30 words
The method of Claim 12, wherein gap errors are assigned in steps (i) and (k) only to those gaps which correspond to speech which did not generate a hypothesized word.
Claim 20
Independent Claim
132 words
A method for determining a best sentence hypothesis corresponding to a sequence of hypothesized words, each having a word error and start and stop times associated therewith, comprising the steps of: (a) comparing each hypothesized word to a grammar specification; (b) determining whether, within the constraints of the grammar, each hypothesized word can be attached to the end of any existing partial sentence hypotheses and identifying such existing partial sentence hypotheses; (c) creating new partial sentence hypotheses by attaching each hypothesized word to the end of an identified partial sentence hypotheses; (d) determining whether any partial sentence hypotheses are complete sentence hypotheses within the grammar specification; (e) calculating a complete sentence hypothesis error for each complete sentence hypothesis; and (f) selecting a complete sentence hypothesis based on the results of step (e).
Claim 21
Independent Claim
132 words
A method for determining a best sentence hypothesis corresponding to a sequence of hypothesized words, each having a word error and start and stop times associated therewith, comprising the steps of: (a) comparing each hypothesized word to a grammar specification; (b) determining whether, within the constraints of the grammar, each hypothesized word can be attached to the end of any existing partial sentence hypotheses and identifying such existing partial sentence hypotheses; (c) creating new partial sentence hypotheses by attaching each hypothesized word to the end of an identified partial sentence hypotheses; (d) determining whether any partial sentence hypotheses are complete sentence hypotheses within the grammar specification; (e) calculating a complete sentence hypothesis error for each complete sentence hypothesis; and (f) selecting a complete sentence hypothesis based on the results of step (e).
Claim 22
Dependent Claim
46 words
The method of Claim 17, wherein the selected complete sentence hypothesis is that complete sentence hypothesis which first has a complete sentence hypothesis error less than a predetermined threshold and which has no subsequent complete sentence hypotheses having complete sentence hypothesis errors less than its own.
Claim 23
Dependent Claim
61 words
The method of Claim 17, wherein the selected complete sentence hypothesis is that complete sentence hypothesis which first has a complete sentence hypothesis error less than a predetermined threshold and which has no subsequent complete sentence hypotheses having complete sentence hypothesis errors less than its own which are generated within a predetermined period after its complete sentence hypothesis error is calculated.
Claim 24
Dependent Claim
47 words
The method of Claim 17, wherein a partial sentence hypothesis error is calculated each time a new partial sentence is hypothesized, and wherein step (e) comprises the step of calculating a partial sentence hypothesis error for the partial sentence hypothesis ending in the newly added hypothesized word.
Claim 25
Dependent Claim
136 words
The method of Claim 20, wherein the partial sentence hypothesis error calculation comprises the steps of: (g) associating a word error with each hypothesized word; (h) associating a gap error with each hypothesized word which begins after the end of the last word in the partial sentence hypothesis identified in step (b); (i) associating an overlap error with each hypothesized word which begins before the end of the last word in the partial sentence hypothesis identified in step (b); (j) summing the errors associated with each hypothesized word in steps (f), (g) and (h) together with the partial sentence error of the partial sentence hypothesis identified in step (b) to give a new partial sentence hypothesis error; and (k) associating the new partial sentence hypothesis error with the new partial sentence hypothesis created in step (c);
Claim 26
Dependent Claim
25 words
The method of Claim 21 wherein the gap error of step (g) is zero whenever silence is associated with the gap between consecutive hypothesized words.
Claim 27
Dependent Claim
33 words
The method of Claim 17, further including the step of: (I) selecting a partial sentence hypothesis to be a best complete sentence hypothesis if a predetermined period elapses wherein no words are hypothesized.
Claim 28
Dependent Claim
26 words
The method of Claim 17, further including the step of: (m) aborting the best sentence determination if a predetermined period elapses wherein no words are hypothesized.
Claim 29
Dependent Claim
22 words
The method of Claim 17, wherein the grammar specification includes states representing transitional utterances commonly occuring between selected pairs of connected words.
Claim 30
Independent Claim
68 words
A method for recognizing connected speech utterances, comprising the steps of: (a) generating a sequence of word hypotheses; (b) comparing each hypothesized word to a grammar specification; (c) generating partial and complete sentence hypotheses by attaching each word hypothesis to the end of a previous partial sentence hypothesis according to the constraints of the grammar specification; and (d) selecting a complete sentence hypothesis having a lowest sentence error.
Claim 31
Independent Claim
68 words
A method for recognizing connected speech utterances, comprising the steps of: (a) generating a sequence of word hypotheses; (b) comparing each hypothesized word to a grammar specification; (c) generating partial and complete sentence hypotheses by attaching each word hypothesis to the end of a previous partial sentence hypothesis according to the constraints of the grammar specification; and (d) selecting a complete sentence hypothesis having a lowest sentence error.
Claim 32
Dependent Claim
37 words
The method of Claim 26, further comprising the step of: (e) when a complete sentence is selected, changing to a second grammar specification, wherein additional partial and complete sentences are generated according to the second grammar specification.
Claim 33
Dependent Claim
20 words
The method of Claim 27, wherein a complete sentence according to the first grammar must be followed by a pause.
References Cited (Backward Citations)
WO1984003983A1
Cited
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.