US5684925A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Speech representation by feature-based word prototypes comprising phoneme targets having reliable high similarity
Assignee: Panasonic Corp of North America
•
Filing: 1995-09-08
•
Granted: 1997-11-04
•
Expires: 2015-09-08
Abstract
Abstract Digitized speech utterances are converted into phoneme similarity data and regions of high similarity are then extracted and used in forming the word prototype. By alignment across speakers unreliable high phoneme similarity regions are eliminated. Word prototype targets are then constructed comprising the following parameters: the phoneme symbol, the average peak height of the phoneme similarity score, the average peak location and the left and right frame locations. For each target a statistical weight is assigned representing the percentage of occurrences the particular high similarity region occurred across all speakers. The word prototype is feature-based allowing a robust speech representation to be constructed without the need for frame-by-frame analysis.
Inventors
- 👤 Philippe R. Morin
- 👤 Ted H. Applebaum
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-06 03:14
CPC Classifications
G; G10; G10L; G10L15/00; G10L15/06; G10L15/063
Claims Analysis
Total Claims: 49
Claim 1
Independent Claim
208 words
A method of electronically representing a given spoken word or phrase as a digital prototype comprising: providing phoneme templates representing a database of standard speech; for a first training instance of said given word or phrase, comparing the first training instance with said phoneme templates to produce first phoneme similarity data as a function of time; processing said first phoneme similarity data to extract first training instance features that exceed a predetermined similarity threshold; for at second training instance of said given word or phrase, comparing the second training instance with said phoneme templates to produce second phoneme similarity data as a function of time; processing said second phoneme similarity data to extract second training instance features that exceed a predetermined similarity threshold; aligning the extracted first and second training instance features and selecting those features that achieve a predetermined degree of correlation between first and second training instances to produce time-dependent speaker-independent phoneme similarity data; building word prototype targets corresponding to the time-dependent speaker-independent phoneme similarity data, the word prototype targets each including a phoneme symbol and at least one datum indicative of a phoneme similarity score; using said phoneme symbol and said phoneme similarity score as a digital prototype to electronically represent the given speech utterance.
Claim 2
Independent Claim
208 words
A method of electronically representing a given spoken word or phrase as a digital prototype comprising: providing phoneme templates representing a database of standard speech; for a first training instance of said given word or phrase, comparing the first training instance with said phoneme templates to produce first phoneme similarity data as a function of time; processing said first phoneme similarity data to extract first training instance features that exceed a predetermined similarity threshold; for at second training instance of said given word or phrase, comparing the second training instance with said phoneme templates to produce second phoneme similarity data as a function of time; processing said second phoneme similarity data to extract second training instance features that exceed a predetermined similarity threshold; aligning the extracted first and second training instance features and selecting those features that achieve a predetermined degree of correlation between first and second training instances to produce time-dependent speaker-independent phoneme similarity data; building word prototype targets corresponding to the time-dependent speaker-independent phoneme similarity data, the word prototype targets each including a phoneme symbol and at least one datum indicative of a phoneme similarity score; using said phoneme symbol and said phoneme similarity score as a digital prototype to electronically represent the given speech utterance.
Claim 3
Dependent Claim
25 words
The method of claim 1 further comprising the step of: determining center height of at least one of said first and second training instance features.
Claim 4
Dependent Claim
25 words
The method of claim 1 further comprising the step of: determining center frame of at least one of said first and second training instance features.
Claim 5
Dependent Claim
26 words
The method of claim 1 further comprising the step of: determining left boundary frame of at least one of said first and second training instance features.
Claim 6
Dependent Claim
26 words
The method of claim 1 further comprising the step of: determining right boundary frame of at least one of said first and second training instance features.
Claim 7
Dependent Claim
34 words
The method of claim 1 further comprising the step of: determining average center height of each of said word prototype targets and storing said average center height in association with said word prototype target.
Claim 8
Dependent Claim
34 words
The method of claim 1 further comprising the step of: determining average center frame of each of said word prototype targets and storing said average center frame in association with said word prototype target.
Claim 9
Dependent Claim
36 words
The method of claim 1 further comprising the step of: determining average left boundary frame of each of said word prototype targets and storing said average left boundary frame in association with said word prototype target.
Claim 10
Dependent Claim
36 words
The method of claim 1 further comprising the step of: determining average right boundary frame of each of said word prototype targets and storing said average right boundary frame in association with said word prototype target.
Claim 11
Dependent Claim
36 words
The method of claim 1 further comprising the step of: determining probability of target observation of each of said word prototype targets and storing said probability of target observation in association with said word prototype target.
Claim 12
Dependent Claim
30 words
The method of claim 1 further comprising the step of: determining weight of each of said word prototype targets and storing said weight in association with said word prototype target.
Claim 13
Dependent Claim
30 words
The method of claim 1 further comprising the step of: determining global statistics of said digital word prototype and storing said global statistics in association with said digital word prototype.
Claim 14
Dependent Claim
43 words
The method of claim 12 wherein said step of determining global statistics of said digital word prototype, further including: determining average number of frames of said digital word prototype and storing said average number of frames in association with said digital word prototype.
Claim 15
Dependent Claim
41 words
The method of claim 12 wherein said step of determining global statistics of said digital word prototype, further including: determining average match rate of said digital word prototype and storing said average match rate in association with said digital word prototype.
Claim 16
Dependent Claim
41 words
The method of claim 12 wherein said step of determining global statistics of said digital word prototype, further including: determining minimum match rate of said digital word prototype and storing said minimum match rate in association with said digital word prototype.
Claim 17
Dependent Claim
41 words
The method of claim 12 wherein said step of determining global statistics of said digital word prototype, further including: determining maximum match rate of said digital word prototype and storing said maximum match rate in association with said digital word prototype.
Claim 18
Dependent Claim
52 words
The method of claim 12 wherein said step of determining global statistics of said digital word prototype, further including: determining average number of detected sections of said first training instance features and said second training instance features and storing said average number of detected sections in association with said digital word prototype.
Claim 19
Dependent Claim
73 words
The method of claim 1 wherein the step ofaligning the extracted first and second training instance features, further including: selecting the extracted first training instance features that correspond to said extracted second training instance features; building an alignment structure that defines a set of aligned peaks between said selected first training instance features and extracted second training instance features and defines data indicative of degree of correlation for each of said aligned peaks.
Claim 20
Dependent Claim
26 words
The method of claim 18 further including: normalizing said selected first training instance features to same time scale as said corresponding extracted second training instance features.
Claim 21
Dependent Claim
25 words
The method of claim 18 further including: performing linear regression between said selected first training instance features and said corresponding extracted first training instance features.
Claim 22
Dependent Claim
45 words
The method of claim 18 further including: normalizing said selected first training instance features to same time scale as said corresponding extracted second training instance features; and performing linear regression between said selected first training instance features and said corresponding extracted second training instance features.
Claim 23
Independent Claim
228 words
An apparatus for electronically representing a given speech utterance signal as a digital word prototype comprising: a phoneme template for representing a database of calibration speech; means for comparing the utterance signal of said first speaker with said phoneme template to produce first speaker phoneme similarity data as a function of time for said given speech utterance signal of a first speaker; means for processing said first speaker phoneme similarity data to extract first speaker features that exceed a predetermined similarity threshold; means for comparing the utterance signal of said second speaker with said phoneme template to produce second speaker phoneme similarity data as a function of time for said given speech utterance signal of a second speaker; means for processing said second speaker phoneme similarity data to extract second speaker features that exceed a predetermined similarity threshold; means for aligning the extracted first and second speaker features and selecting those features that achieve a predetermined degree of correlation between first and second speakers to produce time-dependent speaker-independent phoneme similarity data; means for building word prototype targets corresponding to the time-dependent speaker-independent phoneme similarity data, the word prototype targets each including a phoneme symbol and at least one feature location datum indicative of a time location of that feature; means for using said word prototype targets as said digital word prototype to electronically represent the given speech utterance.
Claim 24
Independent Claim
228 words
An apparatus for electronically representing a given speech utterance signal as a digital word prototype comprising: a phoneme template for representing a database of calibration speech; means for comparing the utterance signal of said first speaker with said phoneme template to produce first speaker phoneme similarity data as a function of time for said given speech utterance signal of a first speaker; means for processing said first speaker phoneme similarity data to extract first speaker features that exceed a predetermined similarity threshold; means for comparing the utterance signal of said second speaker with said phoneme template to produce second speaker phoneme similarity data as a function of time for said given speech utterance signal of a second speaker; means for processing said second speaker phoneme similarity data to extract second speaker features that exceed a predetermined similarity threshold; means for aligning the extracted first and second speaker features and selecting those features that achieve a predetermined degree of correlation between first and second speakers to produce time-dependent speaker-independent phoneme similarity data; means for building word prototype targets corresponding to the time-dependent speaker-independent phoneme similarity data, the word prototype targets each including a phoneme symbol and at least one feature location datum indicative of a time location of that feature; means for using said word prototype targets as said digital word prototype to electronically represent the given speech utterance.
Claim 25
Independent Claim
111 words
A method of electronically representing a given spoken word or phrase as a digital prototype, comprising: providing phoneme templates representing a database of standard speech; providing a plurality of training instances of said given spoken word or phrase; for each training instance, comparing the training instance with said phoneme templates to produce training instance phoneme similarity data as a function of time; building a digital prototype corresponding to the time-dependent phoneme similarity data, said prototype consisting of at least a list of high phoneme similarity region targets, each target having a phoneme identifier and feature data including at least one time location datum indicative of a time location of that phoneme.
Claim 26
Independent Claim
111 words
A method of electronically representing a given spoken word or phrase as a digital prototype, comprising: providing phoneme templates representing a database of standard speech; providing a plurality of training instances of said given spoken word or phrase; for each training instance, comparing the training instance with said phoneme templates to produce training instance phoneme similarity data as a function of time; building a digital prototype corresponding to the time-dependent phoneme similarity data, said prototype consisting of at least a list of high phoneme similarity region targets, each target having a phoneme identifier and feature data including at least one time location datum indicative of a time location of that phoneme.
Claim 27
Dependent Claim
24 words
The method of claim 23 wherein said digital prototype consists of global statistics in addition to said list of high phoneme similarity region targets.
Claim 28
Dependent Claim
28 words
The method of claim 24 further comprising including in the global statistics of said word or phrase prototype the average of the alignment scores of the training instances.
Claim 29
Dependent Claim
27 words
The method of claim 24 further comprising including in the global statistics of said word or phrase prototype the average of the duration of the training instances.
Claim 30
Dependent Claim
31 words
The method of claim 23 further comprising the steps of: producing word prototype targets of said given word or phrase by incrementally merging said phoneme similarity data of said training instances.
Claim 31
Dependent Claim
258 words
The method of claim 27 wherein said merging step is performed as follows: for a first training instance of said given spoken word or phrase, comparing the first training instance with said phoneme templates to produce first phoneme similarity data as a function of time; processing said first phoneme similarity data to extract first training instance high similarity regions for which the phoneme similarity value exceeds a predetermined phoneme similarity threshold; initializing said list of high phoneme similarity region targets as the feature data of the first training instance high similarity regions which exceed said predetermined phoneme similarity threshold; for each succeeding training instance of said given spoken word or phrase, comparing the signal of said succeeding training instance with said phoneme templates to produce succeeding phoneme similarity data as a function of time; processing said succeeding phoneme similarity data to extract succeeding training instance high similarity regions for which the phoneme similarity value exceeds a predetermined phoneme similarity threshold; aligning said succeeding training instance high similarity regions with said list of high phoneme similarity region targets; for each succeeding training instance high similarity region which does not match a high phoneme similarity region target in said alignment, initializing a new high phoneme similarity region target to include the feature data of said matched high similarity region; for each succeeding training instance high similarity region which does match a high phoneme similarity region target in said alignment, updating the values of said high phoneme similarity region target to include the feature data of said matched high similarity region.
Claim 32
Dependent Claim
25 words
The method of claim 23 further comprising the step of: determining the center height of at least one of said training instance high similarity regions.
Claim 33
Dependent Claim
25 words
The method of claim 23 further comprising the step of: determining the center frame of at least one of said training instance high similarity regions.
Claim 34
Dependent Claim
26 words
The method of claim 23 further comprising the step of: determining the left boundary frame of at least one of said training instance high similarity regions.
Claim 35
Dependent Claim
26 words
The method of claim 23 further comprising the step of: determining the right boundary frame of at least one of said training instance high similarity regions.
Claim 36
Dependent Claim
52 words
The method of claim 23 further comprising the step of: determining the average center height of the training instance high similarity regions which match a given high phoneme similarity region target in said alignment, and storing said average center height as feature data in association with said high phoneme similarity region target.
Claim 37
Dependent Claim
52 words
The method of claim 23 further comprising the step of: determining the average center frame of the training instance high similarity regions which match a given high phoneme similarity region target in said alignment, and storing said average center frame as feature data in association with said high phoneme similarity region target.
Claim 38
Dependent Claim
53 words
The method of claim 24 further comprising the step of: determining the average left boundary frame of the training instance high similarity regions which match a given high phoneme similarity region target in said alignment, and storing said average center frame as feature data in association with said high phoneme similarity region target.
Claim 39
Dependent Claim
52 words
The method of claim 24 further comprising the step determining the average right boundary frame of the training instance high similarity regions which match a given high phoneme similarity region target in said alignment, and storing said average center frame as feature data in association with said high phoneme similarity region target.
Claim 40
Dependent Claim
72 words
The method of claim 23 further comprising the step of: determining the probability of a given high phoneme similarity region target as the fraction of said training instances of said given spoken word or phrase containing at least one high similarity region which is matched to said given high phoneme similarity region target in said alignment, and storing said probability as feature data in association with said high phoneme similarity region target.
Claim 41
Dependent Claim
30 words
The method of claim 23 further comprising the step of: determining weight of each of said word prototype targets and storing said weight in association with said word prototype target.
Claim 42
Dependent Claim
30 words
The method of claim 23 further comprising the step of: determining global statistics of said digital word prototype and storing said global statistics in association with said digital word prototype.
Claim 43
Dependent Claim
43 words
The method of claim 39 wherein said step of determining global statistics of said digital word prototype, further including: determining average number of frames of said digital word prototype and storing said average number of frames in association with said digital word prototype.
Claim 44
Dependent Claim
41 words
The method of claim 39 wherein said step of determining global statistics of said digital word prototype, further including: determining average match rate of said digital word prototype and storing said average match rate in association with said digital word prototype.
Claim 45
Dependent Claim
41 words
The method of claim 39 wherein said step of determining global statistics of said digital word prototype, further including: determining minimum match rate of said digital word prototype and storing said minimum match rate in association with said digital word prototype.
Claim 46
Dependent Claim
41 words
The method of claim 39 wherein said step of determining global statistics of said digital word prototype, further including: determining maximum match rate of said digital word prototype and storing said maximum match rate in association with said digital word prototype.
Claim 47
Independent Claim
184 words
An apparatus for electronically representing a given spoken word or phrase as a digital prototype, comprising: a phoneme template for representing a database of standard speech; means for receiving a plurality of training instances of a given spoken word or phrase; means for comparing each training instance of the spoken word or phrase with said phoneme template to produce training instance phoneme similarity data as a function of time for said training instance of the given word or phrase; means for processing said training instance phoneme similarity data to extract regions that exceed a predetermined phoneme similarity threshold; means for producing word prototype targets of said given word or phrase by incrementally merging said training instance phoneme similarity data; means for building target congruence prototypes corresponding to the time-dependent phoneme similarity data, the prototype targets comprising at least a list of phoneme targets, each including a phoneme identifier and feature data including at least one time location datum indicative of a time location of that phoneme; means for using said target congruence prototype as said digital prototype for said given spoken word or phrase.
Claim 48
Independent Claim
184 words
An apparatus for electronically representing a given spoken word or phrase as a digital prototype, comprising: a phoneme template for representing a database of standard speech; means for receiving a plurality of training instances of a given spoken word or phrase; means for comparing each training instance of the spoken word or phrase with said phoneme template to produce training instance phoneme similarity data as a function of time for said training instance of the given word or phrase; means for processing said training instance phoneme similarity data to extract regions that exceed a predetermined phoneme similarity threshold; means for producing word prototype targets of said given word or phrase by incrementally merging said training instance phoneme similarity data; means for building target congruence prototypes corresponding to the time-dependent phoneme similarity data, the prototype targets comprising at least a list of phoneme targets, each including a phoneme identifier and feature data including at least one time location datum indicative of a time location of that phoneme; means for using said target congruence prototype as said digital prototype for said given spoken word or phrase.
Claim 49
Dependent Claim
21 words
The apparatus of claim 44 wherein said prototype targets further comprise global statistics in addition to said list of phoneme targets.
References Cited (Backward Citations)
US3770892A
Cited
US4528688A
Cited
US4481593A
Cited
US4489435A
Cited
US4489434A
Cited
US4624011A
Cited
US4742547A
Cited
US4559602A
Cited
US4723290A
Cited
US4736429A
Cited
US5131043A
Cited
US4780906A
Cited
US5218668A
Cited
US4718094A
Cited
US4908865A
Cited
US4987596A
Cited
US4748670A
Cited
US4820059A
Cited
US4924518A
Cited
US4905287A
Cited
US4803729A
Cited
US5027408A
Cited
US4937871A
Cited
US5197113A
Cited
US5195167A
Cited
US5129001A
Cited
US5345536A
Cited
US5268990A
Cited
US5195168A
Cited
US5369727A
Cited
US5369728A
Cited
US5309547A
Cited
US5241619A
Cited
US5390278A
Cited
US5349645A
Cited
US5233681A
Cited
US5522011A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2000-11-03 | 2001-05-05 | 2001-11-04 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2004-11-03 | 2005-05-05 | 2005-11-04 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2008-11-03 | 2009-05-05 | 2009-11-04 | $7700.00 | Window Closed / Lapsed |