US5864807A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method and apparatus for training a speaker recognition system
Assignee: Google Technology Holdings LLC
•
Filing: 1997-02-25
•
Granted: 1999-01-26
•
Expires: 2017-02-25
Abstract
Abstract A method and apparatus for training a system to assess the identity of a person through the audio characteristics of their voice. The system inserts an audio input (10) into an A/D Converter (20) for processing in a digital signal processor (30). The system then applies Neural network type processing by using a polynomial pattern classifier (60) for training the speaker recognition system.
Inventors
- 👤 William Michael Campbell
- 👤 Khaled Talal Assaleh
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-09 11:27
CPC Classifications
G; G10; G10L; G10L17/00; G10L17/04
Claims Analysis
Total Claims: 31
Claim 1
Independent Claim
238 words
A method for training a speaker recognition system by a computer, comprising steps of: extracting speech parameters from a digitized audio signal to produce a set of differentiating factors (r); storing said set of differentiating factors (r) in a data base to produce a stored set of differentiating factors (r); polynomial pattern classifying by the computer said stored set of differentiating factors (r) to produce a first digital audio signature (w); storing said first digital audio signature (w) in said data base to produce a stored first digital audio signature (w); specifying a first speaker as having audio signature features X 1 , X 2 , . . . , X M ; specifying a second speaker, as having audio signature features Y 1 , Y 2 . . . , Y M ; discriminating between said first speaker and said second speaker; training for said first digital audio signature (w) features of said first speaker, a polynomial for a 2-norm to an ideal output of said first speaker, and an ideal output of 0 for said second speaker; representing a matrix whose rows are a polynomial expansions of said first and second sneakers audio signature features, ##EQU5## and where o 1 is a column vector of length 2M whose first M entries are 1 and remaining entries are 0, and o 2 =1-o 1 and training for said first speaker and said second speaker respectively being: ##EQU6##
Claim 2
Independent Claim
238 words
A method for training a speaker recognition system by a computer, comprising steps of: extracting speech parameters from a digitized audio signal to produce a set of differentiating factors (r); storing said set of differentiating factors (r) in a data base to produce a stored set of differentiating factors (r); polynomial pattern classifying by the computer said stored set of differentiating factors (r) to produce a first digital audio signature (w); storing said first digital audio signature (w) in said data base to produce a stored first digital audio signature (w); specifying a first speaker as having audio signature features X 1 , X 2 , . . . , X M ; specifying a second speaker, as having audio signature features Y 1 , Y 2 . . . , Y M ; discriminating between said first speaker and said second speaker; training for said first digital audio signature (w) features of said first speaker, a polynomial for a 2-norm to an ideal output of said first speaker, and an ideal output of 0 for said second speaker; representing a matrix whose rows are a polynomial expansions of said first and second sneakers audio signature features, ##EQU5## and where o 1 is a column vector of length 2M whose first M entries are 1 and remaining entries are 0, and o 2 =1-o 1 and training for said first speaker and said second speaker respectively being: ##EQU6##
Claim 3
Dependent Claim
39 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein the there is further included a step of providing an audio input of a voice of a speaker from a recorded medium.
Claim 4
Dependent Claim
33 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included a step of converting an audio input to said digitized audio signal.
Claim 5
Dependent Claim
39 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included a step of sending said set of differentiating factors (r) to said data base by a data link.
Claim 6
Dependent Claim
44 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein the step of extracting speech parameters includes a step of determining a frequency domain representation of a short-time power spectra and cepstra of said digitized audio signal.
Claim 7
Dependent Claim
46 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein the step of extracting speech parameters includes a step of determining a frequency domain representation of transitional information (dynamic) short-time power spectra and delta-cepstra of said digitized audio signal.
Claim 8
Dependent Claim
38 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein the step of extracting speech parameters includes a step of determining non-linearly processed filter bank output of said digitized audio signal.
Claim 9
Dependent Claim
35 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein the step of extracting speech parameters includes a step of determining linear-predictive coefficients of the digitized audio signal.
Claim 10
Dependent Claim
52 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein the step of extracting speech parameters resulting in said set of differentiating factors (r) includes a step of adding a new set of said differentiating factors (r) to a stored set of differentiating factors (r).
Claim 11
Dependent Claim
53 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included the step of extracting, storing said set of differentiating factors (r), and polynomial pattern classifying by the computer for a second digitized audio signal to produce a second digital audio signature (w).
Claim 12
Dependent Claim
37 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included the step of comparing said first and second digital audio signatures (w) to recognize a speaker.
Claim 13
Dependent Claim
39 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included the step of permitting communication access by the speaker if first and second digital audio signature (w) correlate.
Claim 14
Dependent Claim
38 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included the step of training said first digital audio signature (w) with said second digital audio signature (w).
Claim 15
Dependent Claim
35 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included the step of storing said second digital audio signature (w) in the data base.
Claim 16
Dependent Claim
76 words
A method for training a speaker recognition system by said computer as claimed in claim 1, wherein there is further included the step of training with said differentiating factors (r), where rows in M correspond to the said first speaker and said speaker audio signature are denoted as M 1 and M 2 respectively: M.sup.t Mw.sub.1 =M.sup.t o.sub.1 (M.sub.1.sup.t M.sub.1 +M.sub.2.sup.t M.sub.2)w.sub.1 =M.sub.1.sup.t 1 (R.sub.1 +R.sub.2)w.sub.1 =M.sub.1.sup.t 1 where 1 is a vector of all ones.
Claim 17
Independent Claim
254 words
An apparatus for training a speaker recognition system comprising: a processor for extracting speech parameters from a digitized audio signal to produce a set of differentiating factors (r); a computer for storing said set of differentiating factors (r) in a data base, said computer coupled to said processor; a polynomial pattern classifier operating on said set of differentiating factors (r) to produce a first digital audio signature (w); means for storing said digitized audio signature (w) in said data base to produce a stored first digital audio signature (w); means for specifying a first speaker as having audio signature features X 1 , X 2 , . . . , X M ; means for specifying a second speaker, as having audio signature features Y 1 , X 2 , . . . , Y M ; means for discriminating between said first speaker and said second speaker; means for training said audio signature of said first and second speaker to provide a polynomial for a 2-norm to an ideal output of 1 for features of said first speaker and an ideal output of 0 for said second speaker; means for representing a matrix whose rows are a polynomial expansions of said first and second speakers audio signature features, ##EQU7## and where o 1 is a column vector of length 2M whose first M entries are 1 and remaining entries are 0, and o 2 =1-o 1 ; and said means for training of said first speaker and said second speaker respectively being: ##EQU8##
Claim 18
Independent Claim
254 words
An apparatus for training a speaker recognition system comprising: a processor for extracting speech parameters from a digitized audio signal to produce a set of differentiating factors (r); a computer for storing said set of differentiating factors (r) in a data base, said computer coupled to said processor; a polynomial pattern classifier operating on said set of differentiating factors (r) to produce a first digital audio signature (w); means for storing said digitized audio signature (w) in said data base to produce a stored first digital audio signature (w); means for specifying a first speaker as having audio signature features X 1 , X 2 , . . . , X M ; means for specifying a second speaker, as having audio signature features Y 1 , X 2 , . . . , Y M ; means for discriminating between said first speaker and said second speaker; means for training said audio signature of said first and second speaker to provide a polynomial for a 2-norm to an ideal output of 1 for features of said first speaker and an ideal output of 0 for said second speaker; means for representing a matrix whose rows are a polynomial expansions of said first and second speakers audio signature features, ##EQU7## and where o 1 is a column vector of length 2M whose first M entries are 1 and remaining entries are 0, and o 2 =1-o 1 ; and said means for training of said first speaker and said second speaker respectively being: ##EQU8##
Claim 19
Dependent Claim
67 words
The apparatus as claimed in claim 16, wherein said polynomial pattern classifier includes means for training with said differentiating factors (r) where rows in M corresponding to said first speaker and said second speaker audio signature are denoted as M 1 and M 2 respectively: M.sup.t Mw.sub.1 =M.sup.t o.sub.1 (M.sub.1.sup.t M.sub.1 +M.sub.2.sup.t M.sub.2)w.sub.1 =M.sub.1.sup.t 1 (R.sub.1 +R.sub.2)w.sub.1 =M.sub.1.sup.t 1 where 1 is a vector of all ones.
Claim 20
Dependent Claim
18 words
The apparatus as claimed in claim 16, wherein there is further included a source of an audio input.
Claim 21
Dependent Claim
36 words
The apparatus as claimed in claim 18, wherein there is further included an analog/digital converter for producing said digitized audio signal from said audio input, said analog/digital converter coupled to said source and to said processor.
Claim 22
Dependent Claim
38 words
The apparatus as claimed in claim 16, wherein there is further included a data link for sending said set of differentiating factors (r) to said data base, said data link coupled to said processor and to said computer.
Claim 23
Dependent Claim
39 words
The apparatus as claimed in claim 16, wherein said processor further includes means for determining a frequency domain representation of a short-time power spectra or cepstra of said digitized audio signal, said means for determining operated by said computer.
Claim 24
Dependent Claim
34 words
The apparatus as claimed in claim 16, wherein said processor further includes means for determining a frequency domain representation of a transitional information (dynamic) short-time power spectra or delta-cepstra of said digitized audio signal.
Claim 25
Dependent Claim
25 words
The apparatus as claimed in claim 16, wherein said processor further includes means for determining non-linearly processed filter bank outputs of said digitized audio signal.
Claim 26
Dependent Claim
22 words
The apparatus as claimed in claim 16, wherein said processor further includes means for determining linear-predictive coefficients of the digitized audio signal.
Claim 27
Dependent Claim
24 words
The apparatus as claimed in claim 16, wherein said processor includes means for extracting speech parameters resulting in said set of differentiating factors (r).
Claim 28
Dependent Claim
25 words
The apparatus as claimed in claim 16, wherein said polynomial pattern classifier includes means for comparing speech parameters in said set of differentiating factors (r).
Claim 29
Dependent Claim
28 words
The apparatus as claimed in claim 16, wherein said polynomial pattern classifier includes means for recognizing speech parameters of a person in said set of differentiating factors (r).
Claim 30
Dependent Claim
42 words
The apparatus as claimed in claim 16, wherein said polynomial pattern classifier includes means for adding a new set of said differentiating factors (r) to a set of differentiating factors (r) already in a data base to train said speaker recognition system.
Claim 31
Dependent Claim
42 words
The apparatus as claimed in claim 16, wherein said polynomial pattern classifier includes means for adding a new set of said differentiating factors (r) to update said set of differentiating factors (r) in a data base to retrain said speaker recognition system.
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) | 2002-01-25 | 2002-07-27 | 2003-01-26 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2006-01-25 | 2006-07-27 | 2007-01-26 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2010-01-25 | 2010-07-27 | 2011-01-26 | $7700.00 | Window Closed / Lapsed |