Patent US5023911A successfully refreshed from Google Patents.
US5023911A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Word spotting in a speech recognition system without predetermined endpoint detection
Assignee: Motorola Solutions Inc
•
Filing: 1988-10-31
•
Granted: 1991-06-11
•
Expires: 2008-10-31
Abstract
Abstract Word spotting in a speech recognition system without predetermining the endpoints of the input speech. The invention is intended to be implemented in a system which has word templates stored in template memory, with the system being capable of accumulating distance measures for states within each word template. The following steps are used to generate a measure of similarity between a subset of the input frames and a word template. The steps are: a) recording a beginning input frame number for each state to identify the potential beginning of the word; b) accumulating distance measures for at least one state for each input frame; c) normalizing the distance measures by substracting a normalization amount from each distance measure; d) recording normalization information corresponding to the normalization amount for each input frame; and e) determining a similarity measure between the word template and a subset of input frames after a given input frame has been processed. The subset is identified from the beginning input frame number corresponding to an end state of the template, through the given input frame number. The similarity measure is based on the normalized distance measure recorded for the end state. and the normalization information.
Inventors
- 👤 Ira A. Gerson
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-26 04:29
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/10; G10L2015/088
Claims Analysis
Total Claims: 25
Claim 1
Independent Claim
343 words
In a speech recognition system, wherein one or more templates each having a plurality of states representing each template are stored for comparing to a series of input frames, said speech recognition system being able to accumulate distance measures for each comparison of the plurality of states to each frame of the series of input frames, a method for identifying a speech utterance from the series of input frames without using predetermined endpoints, comprising the steps of: a) identifying a beginning input frame for a first state of a template to identify the potential beginning of the speech utterance; b) determining a first plurality of distance measures from the comparisons of a first set of states to said beginning input frame; c) determining a second plurality of distance measures from the comparisons of a second set of states to a second input frame; d) determining a first normalization amount for said first plurality of distance measures and determining a second normalization amount for said second plurality of distance measures; e) normalizing said first plurality of distance measures by subtracting said first normalization amount from each distance measure of said first plurality of distance measures and normalizing said second plurality of distance measures by subtracting said second normalization amount from each distance measure of said second plurality of distance measures; f) accumulating each said determined distance measure of said first plurality of distance measures and accumulating each said accumulated distance measure of said second plurality of distance measures in accordance with a corresponding state from said first and second set of states; and g) determining a similarity measure between said template and a subset of the series of input frames, including said beginning input frame, said second input frame, and a third input frame, said subset identified from said beginning input frame through said third input frame, said similarity measure being based on accumulated distance measures corresponding to the plurality of states representing said template and a summation of the normalization amounts including at least said first normalization amount and said second normalization amount.
Claim 2
Dependent Claim
343 words
In a speech recognition system, wherein one or more templates each having a plurality of states representing each template are stored for comparing to a series of input frames, said speech recognition system being able to accumulate distance measures for each comparison of the plurality of states to each frame of the series of input frames, a method for identifying a speech utterance from the series of input frames without using predetermined endpoints, comprising the steps of: a) identifying a beginning input frame for a first state of a template to identify the potential beginning of the speech utterance; b) determining a first plurality of distance measures from the comparisons of a first set of states to said beginning input frame; c) determining a second plurality of distance measures from the comparisons of a second set of states to a second input frame; d) determining a first normalization amount for said first plurality of distance measures and determining a second normalization amount for said second plurality of distance measures; e) normalizing said first plurality of distance measures by subtracting said first normalization amount from each distance measure of said first plurality of distance measures and normalizing said second plurality of distance measures by subtracting said second normalization amount from each distance measure of said second plurality of distance measures; f) accumulating each said determined distance measure of said first plurality of distance measures and accumulating each said accumulated distance measure of said second plurality of distance measures in accordance with a corresponding state from said first and second set of states; and g) determining a similarity measure between said template and a subset of the series of input frames, including said beginning input frame, said second input frame, and a third input frame, said subset identified from said beginning input frame through said third input frame, said similarity measure being based on accumulated distance measures corresponding to the plurality of states representing said template and a summation of the normalization amounts including at least said first normalization amount and said second normalization amount.
Claim 3
Dependent Claim
34 words
The method in accordance with claim 1, further including the step of setting the accumulated distance of one or more initial states of each said template relative to a minimum said accumulated distance measure.
Claim 4
Dependent Claim
25 words
The method in accordance with claim 1, wherein said summation of normalization amounts corresponds to the minimum of said normalized accumulated distances for all states.
Claim 5
Dependent Claim
30 words
The method, in accordance with claim 1 ,wherein the step of determining a similarity measure further includes the step of calculating the number of input frames in said said subset.
Claim 6
Dependent Claim
26 words
The method, in accordance with claim 4, further including the step of calculating an average distance measure per frame for said input frames in said subset.
Claim 7
Dependent Claim
21 words
The method, in accordance with claim 5, further including the step of comparing said average distance to a predetermined distance threshold.
Claim 8
Dependent Claim
20 words
The method, in accordance with claim 4, further including the step of comparing two or more said average distance measures.
Claim 9
Dependent Claim
32 words
The method, in accordance with claim 1, further including the step of determining if the speech energy corresponding to one or more input frames of said subset exceeds a predetermined energy threshold.
Claim 10
Dependent Claim
27 words
The method, in accordance with claim 1, wherein the step of accumulating distance measures further includes the step of comparing said distance measures to a truncating threshold.
Claim 11
Dependent Claim
39 words
The method, in accordance with claim 1, further including the step of outputting a signal indicating that said template matches said subset after it is determined that no other beginning frame number is less than said given frame number.
Claim 12
Dependent Claim
22 words
The method, in accordance with claim 1, wherein the step of recording said beginning input frame number includes recording using modulo arithmetic.
Claim 13
Dependent Claim
20 words
The method, in accordance with claim 1, wherein the step of recording said normalization information includes recording using modulo arithmetic.
Claim 14
Dependent Claim
18 words
The method in accordance with claim 1 wherein said first set of states further includes said first state.
Claim 15
Dependent Claim
18 words
The method in accordance with claim 1 wherein said second set of states further includes said first state.
Claim 16
Dependent Claim
270 words
In a speech recognition system, wherein one or more templates, each having a plurality of states, are stored for comparing to a series of input frames, said speech recognition system being able to accumulate distance measures for each comparison of the plurality of states to each input frame processed, a method for identifying a speech utterance from said series of input frames without using predetermined endpoints, comprising the steps of: a) recording a beginning input frame number for a first state of a template to identify the potential beginning of said speech utterance; b) determining distance measures for each state of the plurality of states of a template for a selected input frame; c) determining a normalization amount for said determined distance measures; d) normalizing said determined distance measures by subtracting said normalization amount from each said determined distance measure; e) accumulating each determined distance measure corresponding to said selected frame for each state of the plurality of states of a template; f) repeating steps b), c), d), and e) for each subsequent selected input frame through a given input frame; and g) determining a similarity measure between said template and a subset of said input frames after said given input frame has been processed, said subset identified from said recorded beginning input frame number through said given input frame, said similarity measure based on said accumulated distance measures of said subset of input frames and corresponding to the plurality of states representing said template and a summation of said normalization amounts including at least a normalization amount for said input frame and a normalization amount for said given frame.
Claim 17
Dependent Claim
270 words
In a speech recognition system, wherein one or more templates, each having a plurality of states, are stored for comparing to a series of input frames, said speech recognition system being able to accumulate distance measures for each comparison of the plurality of states to each input frame processed, a method for identifying a speech utterance from said series of input frames without using predetermined endpoints, comprising the steps of: a) recording a beginning input frame number for a first state of a template to identify the potential beginning of said speech utterance; b) determining distance measures for each state of the plurality of states of a template for a selected input frame; c) determining a normalization amount for said determined distance measures; d) normalizing said determined distance measures by subtracting said normalization amount from each said determined distance measure; e) accumulating each determined distance measure corresponding to said selected frame for each state of the plurality of states of a template; f) repeating steps b), c), d), and e) for each subsequent selected input frame through a given input frame; and g) determining a similarity measure between said template and a subset of said input frames after said given input frame has been processed, said subset identified from said recorded beginning input frame number through said given input frame, said similarity measure based on said accumulated distance measures of said subset of input frames and corresponding to the plurality of states representing said template and a summation of said normalization amounts including at least a normalization amount for said input frame and a normalization amount for said given frame.
Claim 18
Dependent Claim
30 words
The method, in accordance with claim 15, wherein the step of determining a similarity measure further includes the step of calculating the number of input frames in said said subset.
Claim 19
Dependent Claim
24 words
The method, in accordance with claim 15, further including the step of calculating an average distance measure for said input frames in said subset.
Claim 20
Dependent Claim
21 words
The method, in accordance with claim 17, further including the step of comparing said average distance to a predetermined distance threshold.
Claim 21
Dependent Claim
20 words
The method, in accordance with claim 17, further including the step of comparing two or more said average distance measures.
Claim 22
Dependent Claim
32 words
The method, in accordance with claim 15, further including the step of determining if the speech energy corresponding to one or more input frames of said subset exceeds a predetermined energy threshold.
Claim 23
Dependent Claim
27 words
The method, in accordance with claim 15, wherein the step of accumulating distance measures further includes the step of comparing said distance measures to a truncating threshold.
Claim 24
Dependent Claim
39 words
The method, in accordance with claim 15, further including the step of outputting a signal indicating that said template matches said subset after it is determined that no other beginning frame number is less than said given frame number.
Claim 25
Dependent Claim
22 words
The method, in accordance with claim 15, wherein the step of recording said beginning input frame number includes recording using modulo arithmetic.
References Cited (Backward Citations)
US4277644A
Cited
US4319085A
Cited
US4349700A
Cited
US4348553A
Cited
US4400828A
Cited
US4481593A
Cited
US4489435A
Cited
US4590605A
Cited
US4449190A
Cited
US4624010A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1994-06-10 | 1994-12-10 | 1995-06-11 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1998-06-10 | 1998-12-10 | 1999-06-11 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2002-06-10 | 2002-12-10 | 2003-06-11 | $7700.00 | Window Closed / Lapsed |