US4989248A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speaker-dependent connected speech word recognition method
Assignee: Texas Instruments Inc
•
Filing: 1989-03-03
•
Granted: 1991-01-29
•
Expires: 2009-03-03
Abstract
Abstract A cost-effective word recognizer. Each frame of spoken input is compared to a set of reference frames. The comparison is equivalent to embodying the reference frame as an LPC inverse filter, and is preferably done in the autocorrelation domain. To avoid the instability and computational difficulties which can be caused by a high-gain LPC inverse filter, a noise floor is introduced into each reference frame sample. Thus, for each input speech frame, a scalar measures its similarity to each of the vocabulary of reference frames. To achieve connected word recognition based on this similarity measurement, a dynamic programming algorithm is used in which time warping to match a sample to a reference is in effect permitted, and in which matching is performed with unconstrained endpoints. Thus, the word boundary decisions are made on the basis of a local maximum in similarity, and, since no separate word division decision is required, the error which can be introduced by even the best preliminary decision as to word boundaries is avoided.
Inventors
- 👤 Thomas B. Schalk
- 👤 George R. Doddington
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-29 16:37
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/12
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
195 words
A method for recognizing speech comprising the steps of: storing reference templates containing speech parameters corresponding to a plurality of portions of words in a predefined vocabulary and defining speech reference frames; receiving speech inputs and transforming the speech inputs into a plurality of sample input speech frames of speech parameters of like type to those contained in said reference templates; making a sliding comparison of an input speech frame sequence to a reference template sequence; calculating error values indicating the distance between each input speech frame and each reference template sequence corresponding to a word; tracking a scanning error over a predetermined time period; determining when the scanning error has achieved a minimum error of all such comparisons of an input speech frame sequence with each of the reference template sequences corresponding to all of the words in the predefined vocabulary and thereafter increased by a predetermined amount; identifying a local minimum based upon the achievement of a minimum scanning error and the increase therefrom of a predetermined amount; and generating a word recognition hypothesis if the particular value of the accumulated scanning error causing the local minimum is below a certain threshold value.
Claim 2
Independent Claim
195 words
A method for recognizing speech comprising the steps of: storing reference templates containing speech parameters corresponding to a plurality of portions of words in a predefined vocabulary and defining speech reference frames; receiving speech inputs and transforming the speech inputs into a plurality of sample input speech frames of speech parameters of like type to those contained in said reference templates; making a sliding comparison of an input speech frame sequence to a reference template sequence; calculating error values indicating the distance between each input speech frame and each reference template sequence corresponding to a word; tracking a scanning error over a predetermined time period; determining when the scanning error has achieved a minimum error of all such comparisons of an input speech frame sequence with each of the reference template sequences corresponding to all of the words in the predefined vocabulary and thereafter increased by a predetermined amount; identifying a local minimum based upon the achievement of a minimum scanning error and the increase therefrom of a predetermined amount; and generating a word recognition hypothesis if the particular value of the accumulated scanning error causing the local minimum is below a certain threshold value.
Claim 3
Dependent Claim
68 words
A method for recognizing speech as set forth in claim 1, wherein the generation of a word recognition hypothesis occurs when a local minimum in the scanning error sequence has been determined, the local minimum defining a valley point error is less than said certain threshold value, and the difference between the valley point error and the next best word error is greater than a second threshold value.
Claim 4
Dependent Claim
35 words
A method for recognizing speech as set forth in claim 2, further including testing each word recognition hypothesis to ensure that a time period exceeding a third threshold value has elapsed between tentatively identified words.
Claim 5
Dependent Claim
68 words
A method for recognizing speech as set forth in claim 1, wherein speech inputs are received as connected speech during a single utterance in which a plurality of words are stated; and further including generating a plurality of word recognition hypotheses corresponding to the single utterance; and storing each word recognition hypothesis as an identifier indicating a respective word in the predefined vocabulary and a time associated therewith.
Claim 6
Dependent Claim
34 words
A method for recognizing speech as set forth in claim 4, wherein the time associated with each word recognition hypothesis indicates the end of the input speech frame sequence giving rise to the hypothesis.
Claim 7
Dependent Claim
28 words
A method for recognizing speech as set forth in claim 1, wherein the input speech frames have a frame length less than that of said speech reference frames.
Claim 8
Dependent Claim
29 words
A method for recognizing speech as set forth in claim 6, wherein the input speech frames have a frame length which is one-half that of said speech reference frames.
Claim 9
Dependent Claim
88 words
A method for recognizing speech as set forth in claim 1, wherein a local minimum is identified when an input speech frame sequence ending in a current input speech frame gives a calculated error value which is less than the calculated error value for an input speech frame sequence ending with the input speech frame preceding the current input speech frame, and which is less than the calculated error value for the input speech frame sequence ending with the input speech frame following the current input speech frame.
Claim 10
Independent Claim
236 words
A method for recognizing words embedded in connected speech, comprising the steps of: storing reference templates containing speech parameters corresponding to a plurality of portions of words in a predefined vocabulary and defining speech reference frames; receiving speech inputs as connected speech in which a plurality of words are embedded; transforming the speech inputs into a plurality of sample input speech frames of speech parameters of like type of those contained in said reference templates and at a ratio of two sample input speech frames to every speech reference frame; making a sliding comparison of an input speech frame sequence to a reference template sequence; calculating error values indicating the distance between each input speech frame and each of the reference template sequences corresponding to a word; tracking a scanning error over a predetermined time period; identifying a local minimum as a valley in a subsequence of error values which begin and end with error values greater than or equal to the local valley multiplied by a minimum peak-to-valley ratio; generating a word recognition hypothesis when a local valley in the scanning error sequence has been determined; generating additional word recognition hypotheses with respect to the same input of connected speech by repeating the procedure for identifying a local valley; and testing each word recognition hypothesis as so generated to determine that a time period exceeding a predetermined threshold value has elapsed between tentatively identified words.
Claim 11
Independent Claim
236 words
A method for recognizing words embedded in connected speech, comprising the steps of: storing reference templates containing speech parameters corresponding to a plurality of portions of words in a predefined vocabulary and defining speech reference frames; receiving speech inputs as connected speech in which a plurality of words are embedded; transforming the speech inputs into a plurality of sample input speech frames of speech parameters of like type of those contained in said reference templates and at a ratio of two sample input speech frames to every speech reference frame; making a sliding comparison of an input speech frame sequence to a reference template sequence; calculating error values indicating the distance between each input speech frame and each of the reference template sequences corresponding to a word; tracking a scanning error over a predetermined time period; identifying a local minimum as a valley in a subsequence of error values which begin and end with error values greater than or equal to the local valley multiplied by a minimum peak-to-valley ratio; generating a word recognition hypothesis when a local valley in the scanning error sequence has been determined; generating additional word recognition hypotheses with respect to the same input of connected speech by repeating the procedure for identifying a local valley; and testing each word recognition hypothesis as so generated to determine that a time period exceeding a predetermined threshold value has elapsed between tentatively identified words.
Claim 12
Dependent Claim
28 words
A method for recognizing speech as set forth in claim 1, wherein the speech parameters contained in said reference templates and said input speech frames are LPC parameters.
Claim 13
Dependent Claim
32 words
A method for recognizing words embedded in connected speech as set forth in claim 9, wherein the speech parameters contained in said reference templates and said input speech frames are LPC parameters.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1994-01-28 | 1994-07-30 | 1995-01-29 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1998-01-28 | 1998-07-30 | 1999-01-29 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2002-01-28 | 2002-07-30 | 2003-01-29 | $7700.00 | Window Closed / Lapsed |