Patent US6131089A successfully refreshed from Google Patents.
US6131089A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Pattern classifier with training system and methods of operation therefor
Assignee: Google Technology Holdings LLC
•
Filing: 1998-05-04
•
Granted: 2000-10-10
•
Expires: 2018-05-04
Abstract
Abstract Classifiers (110) and a comparator (112) perform an identification method (400) to identify a class as one of a predetermined set of classes. The identification method is based on determining the observation costs associated with the unidentified class. The identification method includes combining models representing the predetermined set of classes and the unidentified vectors representing the class. The predetermined class associated with the largest observation cost is identified as the class. Additionally, a unique, low-complexity training method (300) includes creating the models which represent the predetermined set of classes.
Inventors
- 👤 William Michael Campbell
- 👤 Bruce Alan Fette
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 15:44
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/08; G10L15/18
Claims Analysis
Total Claims: 27
Claim 1
Independent Claim
229 words
A method for training a set of models by classifier and training system, each of the set of models representing at least part of a predetermined speech recognition class, the predetermined class being one of a set of predetermined classes, the method comprising the steps of: associating vectors for the set of predetermined classes with at least one of a group of predetermined states, each of the group of predetermined states representing at least one of the set of models; combining the vectors to determine an individual model structure for each of the set of models; producing a combined model structure for each of the set of models based on the individual model structure; and creating each of the set of models based on the combined model structure and the vectors, the method identifying a class as at least one of the set of predetermined classes, wherein the method further comprises the steps of: determining unidentified vectors which represent the class; multiplying selected ones of the set of models with the unidentified vectors to determine a cost associated with each of the unidentified vectors; accumulating the cost for each of the unidentified vectors to determine a total cost for the unidentified vectors; and identifying the speech recognition class by the classifier and training system as at least one of the set of predetermined classes based on the total cost.
Claim 2
Independent Claim
229 words
A method for training a set of models by classifier and training system, each of the set of models representing at least part of a predetermined speech recognition class, the predetermined class being one of a set of predetermined classes, the method comprising the steps of: associating vectors for the set of predetermined classes with at least one of a group of predetermined states, each of the group of predetermined states representing at least one of the set of models; combining the vectors to determine an individual model structure for each of the set of models; producing a combined model structure for each of the set of models based on the individual model structure; and creating each of the set of models based on the combined model structure and the vectors, the method identifying a class as at least one of the set of predetermined classes, wherein the method further comprises the steps of: determining unidentified vectors which represent the class; multiplying selected ones of the set of models with the unidentified vectors to determine a cost associated with each of the unidentified vectors; accumulating the cost for each of the unidentified vectors to determine a total cost for the unidentified vectors; and identifying the speech recognition class by the classifier and training system as at least one of the set of predetermined classes based on the total cost.
Claim 3
Dependent Claim
43 words
A method as claimed in claim 1, further comprising the steps of: vector quantizing coefficients for the vectors for each of the set of models; and performing a polynomial expansion of the coefficients for each of the vectors before performing the combining step.
Claim 4
Dependent Claim
51 words
A method as claimed in claim 1, further comprising the step of determining a total model structure based on the individual model structure for each of the set of models, and wherein the combined model structure for each of the set of models is further based on the total model structure.
Claim 5
Dependent Claim
58 words
A method as claimed in claim 1, further comprising the steps of: mapping the combined model structure to a matrix for each of the set of models; and decomposing the matrix for each of the set of models, and wherein the creating step includes determining each of the set of models based on the matrix and the vectors.
Claim 6
Dependent Claim
29 words
A method as claimed in claim 1, further including the step of performing a polynomial expansion of coefficients for each of the unidentified vectors before performing the multiplying step.
Claim 7
Dependent Claim
18 words
A method as claimed in claim 1, wherein each of the group of predetermined states represents a phoneme.
Claim 8
Dependent Claim
19 words
A method as claimed in claim 1, wherein each of the set of predetermined classes represents a spoken word.
Claim 9
Dependent Claim
19 words
A method as claimed in claim 1, wherein each of the set of predetermined classes represents a digital image.
Claim 10
Dependent Claim
19 words
A method as claimed in claim 1, wherein each of the set of predetermined classes represents a radio signature.
Claim 11
Dependent Claim
18 words
A method as claimed in claim 1, wherein each of the set of predetermined classes represents a speaker.
Claim 12
Independent Claim
122 words
A method for identifying a speech recognition class by classifier and training system as at least one of a set of predetermined classes, each of the set of predetermined classes being represented by at least one of a set of models, the method comprising the steps of: determining unidentified vectors which represent the class; multiplying selected ones of the set of models with the unidentified vectors to determine a cost associated with each of the unidentified vectors; accumulating the cost for each of the unidentified vectors to determine a total cost for the unidentified vectors; and identifying the speech recognition class by the classifier and training system as at least one of the set of predetermined classes based on the total cost.
Claim 13
Independent Claim
122 words
A method for identifying a speech recognition class by classifier and training system as at least one of a set of predetermined classes, each of the set of predetermined classes being represented by at least one of a set of models, the method comprising the steps of: determining unidentified vectors which represent the class; multiplying selected ones of the set of models with the unidentified vectors to determine a cost associated with each of the unidentified vectors; accumulating the cost for each of the unidentified vectors to determine a total cost for the unidentified vectors; and identifying the speech recognition class by the classifier and training system as at least one of the set of predetermined classes based on the total cost.
Claim 14
Dependent Claim
29 words
A method as claimed in claim 11, further including the step of performing a polynomial expansion of coefficients for each of the unidentified vectors before performing the multiplying step.
Claim 15
Dependent Claim
129 words
A method as claimed in claim 11, for training the set of models, each of the set of models representing at least part of a predetermined class, the predetermined class being one of the set of predetermined classes, the method comprising the steps of: associating vectors for the set of predetermined classes with at least one of a group of predetermined states, each of the group of predetermined states representing at least one of the set of models; combining the vectors to determine an individual model structure for each of the set of models; producing a combined model structure for each of the set of models based on the individual model structure; and creating each of the set of models based on the combined model structure and the vectors.
Claim 16
Dependent Claim
43 words
A method as claimed in claim 13, further comprising the steps of: vector quantizing coefficients for the vectors for each of the set of models; and performing a polynomial expansion of the coefficients for each of the vectors before performing the combining step.
Claim 17
Dependent Claim
51 words
A method as claimed in claim 13, further comprising the step of determining a total model structure based on the individual model structure for each of the set of models, and wherein the combined model structure for each of the set of models is further based on the total model structure.
Claim 18
Dependent Claim
58 words
A method as claimed in claim 13, further comprising the steps of: mapping the combined model structure to a matrix for each of the set of models; and decomposing the matrix for each of the set of models, and wherein the creating step includes determining each of the set of models based on the matrix and the vectors.
Claim 19
Dependent Claim
186 words
A classifier and training system for identifying a speech recognition class as at least one of a set of predetermined classes, the class being represented by a plurality of unidentified vectors, each of the set of predetermined classes being represented by at least one of a set of predetermined models, each of the set of predetermined models representing a predetermined state, the predetermined state being one of a group of predetermined states, the system comprising: a plurality of classifiers for receiving the set of predetermined models and the plurality of unidentified vectors and generating costs, wherein each of the plurality of classifiers is further comprised of: a set of model multipliers for receiving models and unidentified vectors to generate the costs; a selector for receiving the costs, enabling selected ones of the set of model multipliers based on the costs, and storing the costs in a memory; and a comparator coupled to each of the plurality of classifiers for comparing the costs generated from each of the plurality of classifiers and identifying the speech recognition class by the classifier and training system based on the costs.
Claim 20
Dependent Claim
186 words
A classifier and training system for identifying a speech recognition class as at least one of a set of predetermined classes, the class being represented by a plurality of unidentified vectors, each of the set of predetermined classes being represented by at least one of a set of predetermined models, each of the set of predetermined models representing a predetermined state, the predetermined state being one of a group of predetermined states, the system comprising: a plurality of classifiers for receiving the set of predetermined models and the plurality of unidentified vectors and generating costs, wherein each of the plurality of classifiers is further comprised of: a set of model multipliers for receiving models and unidentified vectors to generate the costs; a selector for receiving the costs, enabling selected ones of the set of model multipliers based on the costs, and storing the costs in a memory; and a comparator coupled to each of the plurality of classifiers for comparing the costs generated from each of the plurality of classifiers and identifying the speech recognition class by the classifier and training system based on the costs.
Claim 21
Independent Claim
126 words
A method for identifying a speech recognition class by classifier and training system as at least one of a set of predetermined classes, the method comprising the steps of: representing each of the set of predetermined classes by at least one of a set of models; training the set of models; determining unidentified vectors which represent the class; multiplying selected ones of the set of models with the unidentified vectors to determine a cost associated with each of the unidentified vectors; accumulating the cost for each of the unidentified vectors to determine a total cost for the unidentified vectors; and identifying the speech recognition class by the classifier and training system as at least one of the set of predetermined classes based on the total cost.
Claim 22
Independent Claim
126 words
A method for identifying a speech recognition class by classifier and training system as at least one of a set of predetermined classes, the method comprising the steps of: representing each of the set of predetermined classes by at least one of a set of models; training the set of models; determining unidentified vectors which represent the class; multiplying selected ones of the set of models with the unidentified vectors to determine a cost associated with each of the unidentified vectors; accumulating the cost for each of the unidentified vectors to determine a total cost for the unidentified vectors; and identifying the speech recognition class by the classifier and training system as at least one of the set of predetermined classes based on the total cost.
Claim 23
Dependent Claim
29 words
A method as claimed in claim 18, further including the step of performing a polynomial expansion of coefficients for each of the unidentified vectors before performing the multiplying step.
Claim 24
Dependent Claim
101 words
A method as claimed in claim 18, wherein the training step further includes the steps of: associating vectors for the set of predetermined classes with at least one of a group of predetermined states, each of the group of predetermined states representing at least one of the set of models; combining the vectors to determine an individual model structure for each of the set of models; producing a combined model structure for each of the set of models based on the individual model structure; and creating each of the set of models based on the combined model structure and the vectors.
Claim 25
Dependent Claim
43 words
A method as claimed in claim 20, further comprising the steps of: vector quantizing coefficients for the vectors for each of the set of models; and performing a polynomial expansion of the coefficients for each of the vectors before performing the combining step.
Claim 26
Dependent Claim
51 words
A method as claimed in claim 20, further comprising the step of determining a total model structure based on the individual model structure for each of the set of models, and wherein the combined model structure for each of the set of models is further based on the total model structure.
Claim 27
Dependent Claim
58 words
A method as claimed in claim 20, further comprising the steps of: mapping the combined model structure to a matrix for each of the set of models; and decomposing the matrix for each of the set of models, and wherein the creating step includes determining each of the set of models based on the matrix and the vectors.
References Cited (Backward Citations)
US5381513A
Cited
US5751904A
Cited
US5317673A
Cited
US5390136A
Cited
US5509103A
Cited
US5749072A
Cited
US5734793A
Cited
US5854855A
Cited
US5818963A
Cited
US5594834A
Cited
US5638486A
Cited
US5719692A
Cited
US5724486A
Cited
US5895447A
Cited
US5864807A
Cited
US5946653A
Cited
US6038535A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2003-10-10 | 2004-04-10 | 2004-10-10 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2007-10-10 | 2008-04-10 | 2008-10-10 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2011-10-10 | 2012-04-10 | 2012-10-10 | $7700.00 | Window Closed / Lapsed |