US5915236A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Word recognition system which alters code executed as a function of available computational resources
Assignee: Nuance Communications Inc
•
Filing: 1997-06-26
•
Granted: 1999-06-22
•
Expires: 2017-06-26
Abstract
Abstract A word recognition system detects the computational resources available to it, such as the speed or number of processors, or whether there is DSP hardware, and alters the instructions it executes in response. The system can be a word recognition program designed to run on different computers having different computational resources. The program receives user generated word signals representing words to be recognized; performs pattern matching on them to select which vocabulary words most probably correspond to such word signals; detects if certain computational resources are available; and varies the instructions it executes in response. In many embodiments the system is a speech recognition program. The word recognition program can vary the computational intensity of its signal processing as a function of available computational resources. Preferably it can match the same word models against representations of word signals produced both by its more and less intensive signal processing. For example, the more intense signal processing might take more frequent FFTs, but average them over the same time period at which the less intense signal processing takes FFTs, so both can represent word signals with the same data structure. If the program detects both a CPU and a DSP processor, it can cause the DSP to determine when the program should interrupt the CPU. The program can also vary the rate at which it filters relatively low scoring words out of consideration during the recognition process as a function of the level of available computational resources.
Inventors
- 👤 Joel M. Gould
- 👤 Frank J. McGrath
- 👤 Joel W. Parke
- 👤 Dean G. Sturtevant
- 👤 Jed M. Roberts
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 15:02
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/06; G10L15/063; G09; G09B; G09B19/00; G09B19/04; G10L15/08; G10L15/18; G10L15/22; G10L15/28; G10L15/285; G10L15/1815; G10L2015/0631; G10L2015/0638; G10L2015/226; G10L2015/228
Claims Analysis
Total Claims: 18
Claim 1
Independent Claim
121 words
A word recognition program product designed to run on different computer systems having different computational resources available for use by said program product, said program product comprising: means for receiving user generated word signals representing words to be recognized; recognition means for performing pattern matching upon the word signals to select which one or more of a plurality of vocabulary words, according to said pattern matching, most probably correspond to each such word signal; means for automatically detecting if certain computational resources are available for use by said program product; and means for varying the computer instructions executed by said program product in response to whether or not said means for detecting detects such resources are available to the program product.
Claim 2
Dependent Claim
121 words
A word recognition program product designed to run on different computer systems having different computational resources available for use by said program product, said program product comprising: means for receiving user generated word signals representing words to be recognized; recognition means for performing pattern matching upon the word signals to select which one or more of a plurality of vocabulary words, according to said pattern matching, most probably correspond to each such word signal; means for automatically detecting if certain computational resources are available for use by said program product; and means for varying the computer instructions executed by said program product in response to whether or not said means for detecting detects such resources are available to the program product.
Claim 3
Dependent Claim
49 words
A word recognition program product as in claim 1 wherein; said word recognition program product is a speech recognition program product; said means for receiving user generated word signals includes means for receiving acoustic signals representing spoken words; and said recognition means perform speech recognition upon said acoustic signals.
Claim 4
Dependent Claim
91 words
A speech recognition program product as in claim 2 wherein: said recognition means including signal processing means for performing signal processing upon said received acoustic signals to produce processed representations of said received acoustic signals which are used in said pattern matching; and wherein said means for varying computer instructions includes means for causing said signal processing means to execute instructions which cause said signal processing to be either more or less computationally intensive, respectively, in response to a detection that said certain computational resources are or are not are available.
Claim 5
Dependent Claim
86 words
A speech recognition program product as in claim 3: further including means for storing word models in association with said vocabulary words, each of which represents the processed representations of acoustic signals associated with its associated vocabulary word; wherein said recognition means includes means for comparing said processed representations of acoustic signals with said word models to perform said pattern matching; and said means for comparing compares the processed representations produced by both said more and less computationally intensive signal processing against the same word models.
Claim 6
Dependent Claim
112 words
A speech recognition program product as in claim 4 wherein: said less computationally intensive signal processing includes repeatedly performing FFTs upon said acoustic signal at a first frequency to produce a set of spectral energies at different frequencies once during each first frequency period; and said more computationally intensive signal processing includes repeatedly performing FFTs upon said acoustic signal at a second frequency which is a given multiple of said first frequency and then averaging the results of the FFTs during each first frequency period to produce a set of spectral energies once during each such first frequency period which corresponds in form to that produced by said less intensive signal processing.
Claim 7
Dependent Claim
63 words
A speech recognition program product as in claim 3 wherein: said speech recognition program product is designed to run at least in part on one or more central processing units, or CPUs; and said means for automatically detecting includes means for detecting if said computer system includes a certain type of digital signal processor, or DSP, available for use by said program product.
Claim 8
Dependent Claim
102 words
A speech recognition program product as in claim 6 further including: means, responsive to the detection of said certain type of DSP, for instructing said DSP to: perform a detection function on said acoustic signals to determine when one or more of said acoustic signals appears to correspond to one or more words to be recognized; and communicate to said one or more CPUs when such a determination is made; and means running on one or more of said CPUs for responding to said communication by causing said speech recognition means to perform pattern matching upon said one or more acoustic signals.
Claim 9
Dependent Claim
193 words
A word recognition program product as in claim 1 wherein: said means for automatically detecting provides an indication of the computational power available for use by said program product; said recognition means includes: means for storing a plurality of word models, one for each of said vocabulary words; means for comparing and scoring a word model against a given word signal in successive stages, with at least a part of each of a plurality of word models being compared against at least a part of said word signal and being assigned a score representing the closeness of the match between that word model and that word signal in each such stage before advancing to the next stage; and threshold means for preventing a word signal from being compared against word models in subsequent stages of the comparison against a given word signal when its score against that word signal at a given state is worse than a certain threshold measure; and said means for varying the computer instructions executed includes means for varying said threshold measure as a function of said indication of the computational power available for use by said program product.
Claim 10
Dependent Claim
107 words
A computerized word recognition method designed to run on different computer systems having different computational resources available for use by said method, comprising the steps of: receiving user generated word signals representing words to be recognized; performing pattern matching upon the word signals to select which one or more of a plurality of vocabulary words, according to said pattern matching, most probably correspond to each such word signal; automatically detecting if certain computational resources are available for use by said method; and varying the computer instructions executed by said method in response to whether or not said detecting step detects such resources are available to the method.
Claim 11
Dependent Claim
107 words
A computerized word recognition method designed to run on different computer systems having different computational resources available for use by said method, comprising the steps of: receiving user generated word signals representing words to be recognized; performing pattern matching upon the word signals to select which one or more of a plurality of vocabulary words, according to said pattern matching, most probably correspond to each such word signal; automatically detecting if certain computational resources are available for use by said method; and varying the computer instructions executed by said method in response to whether or not said detecting step detects such resources are available to the method.
Claim 12
Dependent Claim
43 words
A computerized word recognition method as in claim 9 wherein; said method is a speech recognition method; said receiving of user generated word signals includes receiving acoustic signals representing spoken words; and said pattern matching is speech recognition performed upon said acoustic signals.
Claim 13
Dependent Claim
75 words
A computerized speech recognition method as in claim 10 wherein: said pattern matching step performs signal processing upon said received acoustic signals to produce processed representations of said received acoustic signals which are used in said pattern matching; and wherein said varying of computer instructions includes causing said signal processing to be either more or, less computationally intensive, respectively, in response to a detection that said certain computational resources are or, are not are available.
Claim 14
Dependent Claim
85 words
A computerized speech recognition method as in claim 11: further including the step of storing word models in association with said vocabulary words, each of which represents the processed representations of acoustic signals associated with its associated vocabulary word; wherein said pattern matching step includes comparing said processed representations of acoustic signals with said word models to perform said pattern matching; and said comparing step compares the processed representations produced by both said more and less computationally intensive signal processing against the same word models.
Claim 15
Dependent Claim
112 words
A computerized speech recognition method as in claim 12 wherein: said less computationally intensive signal processing includes repeatedly performing FFTs upon said acoustic signal at a first frequency to produce a set of spectral energies at different frequencies once during each first frequency period; and said more computationally intensive signal processing includes repeatedly performing FFTs upon said acoustic signal at a second frequency which is a given multiple of said first frequency and then averaging the results of the FFTs during each first frequency period to produce a set of spectral energies once during each such first frequency period which corresponds in form to that produced by said less intensive signal processing.
Claim 16
Dependent Claim
57 words
A computerized speech recognition method as in claim 11 wherein: said method is designed to run at least in part on one or more central processing units, or CPUs; and said step of automatically detecting includes detecting if said computer system includes a certain type of digital signal processor, or DSP, available for use by said method.
Claim 17
Dependent Claim
103 words
A computerized speech recognition method as in claim 14 further including: a step of responding to the detection of said certain type of DSP, by instructing said DSP to: perform a detection function on said acoustic signals to determine when one or more of said acoustic signals appears to correspond to one or more words to be recognized; and communicate to said one or more CPUs when such a determination is made; and a step performed on one or more of said CPUs for responding to said communication by causing said pattern matching to be performed upon said one or more acoustic signals.
Claim 18
Dependent Claim
184 words
A computerized speech recognition method as in claim 9 wherein: said step of automatically detecting provides an indication of the computational power available for use by said method; said step of performing pattern matching includes: storing a plurality of word models, one for each of said vocabulary words; comparing and scoring a word model against a given word signal in successive stages, with at least a part of each of a plurality of word models being compared against at least a part of said word signal and being assigned a score representing the closeness of the match between that word model and that word signal in each such stage before advancing to the next stage; and preventing a word signal from being compared against word models in subsequent stages of the comparison against a given word signal when its score against that word signal at a given state is worse than a certain threshold measure; and said varying of the computer instructions executed includes varying said threshold measure as a function of said indication of the computational power available for use by said method.
References Cited (Backward Citations)
US4297528A
Cited
US4336421A
Cited
US4831653A
Cited
US4394538A
Cited
US4509133A
Cited
US4439161A
Cited
US4651289A
Cited
US4720863A
Cited
US4677673A
Cited
US4624008A
Cited
US4731845A
Cited
US4975959A
Cited
US5003603A
Cited
US4977599A
Cited
US4833713A
Cited
US4751737A
Cited
US4783803A
Cited
US4829575A
Cited
US4776016A
Cited
US4903305A
Cited
US4866778A
Cited
US4903306A
Cited
US4837831A
Cited
US4829576A
Cited
US4837830A
Cited
US5123086A
Cited
US5231670A
Cited
US5065431A
Cited
US5146503A
Cited
US4964077A
Cited
US4979213A
Cited
US5122972A
Cited
US5513289A
Cited
US5027406A
Cited
US4994983A
Cited
US5377303A
Cited
US5036539A
Cited
US5231691A
Cited
US5136654A
Cited
US5097509A
Cited
US5293451A
Cited
US5233251A
Cited
US5386494A
Cited
US5425128A
Cited
US5386492A
Cited
US5428707A
Cited
US5384892A
Cited
US5390279A
Cited
EP0618565A2
Cited
US5465317A
Cited
US5640490A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2002-06-21 | 2002-12-21 | 2003-06-22 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2006-06-21 | 2006-12-21 | 2007-06-22 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2010-06-21 | 2010-12-21 | 2011-06-22 | $7700.00 | Window Closed / Lapsed |