US5805771A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Automatic language identification method and system
Assignee: Texas Instruments Inc
•
Filing: 1994-06-22
•
Granted: 1998-09-08
•
Expires: 2014-06-22
Abstract
Abstract This invention consists of three enhancements to HMM-based automatic language identification systems. The three enhancements are: (i) language-discriminant acoustic model training and recognition, (ii) an acoustic model pruning procedure that picks only those phonetic models which are considered useful for language identification, and (iii) a neural network-based language classification method that uses knowledge-based features derived from phone sequences output by the HMM phonetic recognizers.
Inventors
- 👤 Yeshwant K. Muthusamy
- 👤 Lorin P. Netsch
- 👤 Periagaram K. Rajasekaran
- 👤 Barbara J. Wheatley
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 19:03
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/005; G10L15/08; G10L15/14; G10L15/142; G10L15/144
Claims Analysis
Total Claims: 19
Claim 1
Independent Claim
119 words
A method of automatically identifying a language among a plurality of languages comprising the steps of: forming models for each of the phonetic elements of the speech in each individual language to be identified; after models for a language are created, training the models using a speech corpus containing samples of speech from that language to form within-language models; training models for each language using the speech corpora and models for said plurality of languages except the language used to create the original within-language models to form cross-language models; determining acoustical features for each state of each model to optimally discriminate within-language acoustic distribution from cross-language distribution; and identifying a language from input speech based on said acoustical features.
Claim 2
Independent Claim
119 words
A method of automatically identifying a language among a plurality of languages comprising the steps of: forming models for each of the phonetic elements of the speech in each individual language to be identified; after models for a language are created, training the models using a speech corpus containing samples of speech from that language to form within-language models; training models for each language using the speech corpora and models for said plurality of languages except the language used to create the original within-language models to form cross-language models; determining acoustical features for each state of each model to optimally discriminate within-language acoustic distribution from cross-language distribution; and identifying a language from input speech based on said acoustical features.
Claim 3
Dependent Claim
19 words
The method of claim 1 wherein said forming step includes forming Hidden Markov Models for phones for each language.
Claim 4
Dependent Claim
39 words
The method of claim 2 wherein said Hidden Markov Models structure consists of states and transitions between states where each set of states that are vertically aligned share an acoustical statistical distribution which is a single multivariate Gaussian density.
Claim 5
Dependent Claim
35 words
The method of claim 2 wherein the step of training within-language models includes training using maximum-likelihood estimation procedures to produce a set of phonetic element HMMs representing statistics of the phonetic elements of each language.
Claim 6
Dependent Claim
47 words
The method of claim 4 wherein said step of training to form cross-language models includes the step of discriminative training using within-language HMMs for each language, to produce both within-language and corresponding cross-language models and to provide corresponding acoustical statistical distributions for the within-language and cross-language models.
Claim 7
Dependent Claim
39 words
The method of claim 5 wherein said determining step includes diagonalization of said statistical distributions to determine acoustical features for each state of each model to optimally discriminate the within-language model statistical distribution from the cross-language model statistical distribution.
Claim 8
Dependent Claim
53 words
The method of claim 6 wherein the identifying step includes the step of determining the likelihood that unknown speech is from each language by the steps of; processing input speech in an unknown language in a recognizer for each language and determining likelihood that speech came from the phonetic elements of each language.
Claim 9
Dependent Claim
30 words
The method of claim 7 wherein said calculating step includes the step of providing a set of maximum-likelihood HMM recognizers, each using the HMMs and optimal features for each language.
Claim 10
Dependent Claim
36 words
The method of claim 6 further including the step of pruning the set of HMM phonetic models used in determining the language identification likelihood from the statistics of the resulting likelihood scores for each phonetic HMM.
Claim 11
Dependent Claim
24 words
The method of claim 7 wherein said identifying step includes the step of determining features based on sequences of phones output by the recognizers.
Claim 12
Dependent Claim
35 words
The method of claim 10 wherein said features include frequency of occurrence of pairs and triples of phones, duration statistics of phones, phone-pairs and phone-triples, ratios of occurrence of phones with respect to each other.
Claim 13
Dependent Claim
34 words
The method of claim 11 wherein said identifying step includes the step of feeding the features computed on the sequence of phones output by the recognizers into a neural network that performs language classification.
Claim 14
Independent Claim
72 words
A system for identifying an input natural language comprising: a phonetic recognizer responsive to input speech for providing sequences of phones for a plurality of natural languages; means coupled to said recognizer for knowledge-based selection and computation of features specifically designed to distinguish between natural languages based on said sequences of phones; and a neural network responsive to said features for identifying a natural language from among a plurality of natural languages.
Claim 15
Independent Claim
72 words
A system for identifying an input natural language comprising: a phonetic recognizer responsive to input speech for providing sequences of phones for a plurality of natural languages; means coupled to said recognizer for knowledge-based selection and computation of features specifically designed to distinguish between natural languages based on said sequences of phones; and a neural network responsive to said features for identifying a natural language from among a plurality of natural languages.
Claim 16
Independent Claim
164 words
A method of automatically identifying a language among a plurality of languages comprising the steps of: forming models for each of the phonetic elements of the speech in each individual language to be identified; after models for a language are created, training the models using a speech corpus containing samples of speech from that language to form within-language model; training models for each language using the speech corpora and models for said plurality of languages except the language used to create the original within-language models to form cross-language models; determining acoustical features for each state of each model to optimally discriminate within-language acoustic distribution from cross-language distribution; and identifying a language from input speech based on said acoustical features wherein said identifying step includes said step of determining the likelihood that unknown speech is from each language by processing input speech in an unknown language in a recognizer for each language and determining likelihood that speech came from the phonetic elements of each language.
Claim 17
Independent Claim
164 words
A method of automatically identifying a language among a plurality of languages comprising the steps of: forming models for each of the phonetic elements of the speech in each individual language to be identified; after models for a language are created, training the models using a speech corpus containing samples of speech from that language to form within-language model; training models for each language using the speech corpora and models for said plurality of languages except the language used to create the original within-language models to form cross-language models; determining acoustical features for each state of each model to optimally discriminate within-language acoustic distribution from cross-language distribution; and identifying a language from input speech based on said acoustical features wherein said identifying step includes said step of determining the likelihood that unknown speech is from each language by processing input speech in an unknown language in a recognizer for each language and determining likelihood that speech came from the phonetic elements of each language.
Claim 18
Dependent Claim
24 words
The method of claim 14 wherein said identifying step includes the step of determining features based on sequences of phones output by the recognizers.
Claim 19
Dependent Claim
35 words
The method of claim 15 wherein said features include frequency of occurrence of pairs and triples of phones, duration statistics of phones, phone-pairs and phone-triples, ratios of occurrence of phones with respect to each other.
References Cited (Backward Citations)
US4829577A
Cited
US5150449A
Cited
US5175793A
Cited
US5278911A
Cited
US5307444A
Cited
US5479563A
Cited
US5241619A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2001-09-07 | 2002-03-09 | 2002-09-08 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2005-09-07 | 2006-03-09 | 2006-09-08 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2009-09-07 | 2010-03-09 | 2010-09-08 | $7700.00 | Window Closed / Lapsed |