Patent EP0424071A2 successfully refreshed from Google Patents.
EP0424071A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Speaker recognition
Assignee: Logica UK Ltd
•
Filing: 1990-10-15
•
Granted: —
•
Expires: 2010-10-15
Abstract
Abstract A speaker recognition system tests data during enrolment by preparing a template for a word from repeated utterances, comparing further repeated utterances with the template and updating the template with accepted utterances, and finally accepting or rejecting the updated template on the basis of how many of the further utterances are accepted and how many rejected. This protects the quality of the template formed from the enrolment utterances. In addition to storing final versions of the templates for each of a plurality of words, data is stored relating to the co-variance of utterances of different words. During analysis of the spoken inputs, the pitch of voiced sections is determined and this is used both to enable determination of the start and end points of the spoken utterance and also to determine the timing of analysis frame periods. During speaker recognition, a plurality of test utterances are obtained from the unknown speaker, and these are analysed separately and the analysis results combined using the co-variance data to make a recognition decision. The system is particularly useful for speaker verification to verify a previous claimed identity. The system has particular application to the recognition of individuals over the telephone.
Inventors
- 👤 Jeremy Burford Peckham
- 👤 Eleftherios Dimitrios Frangoulis
- 👤 Trevor Joseph Thomas
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-09 13:29
CPC Classifications
G; G10; G10L; G10L17/00; G10L17/04; G10L25/00; G10L25/78
Claims Analysis
Total Claims: 33
Claim 1
Independent Claim
83 words
A method of enrolling speakers in a speech analysis system in which a speaker being enrolled provides a plurality of utterances at least one of which is subjected to a comparison operation based on one or more others of the plurality of utterances, and in accordance with the results of the comparison the system decides whether to reject or accept a said utterance, the system storing characteristics of the speech input to it by the speaker derived from analysis only of non-rejected utterances.
Claim 2
Independent Claim
83 words
A method of enrolling speakers in a speech analysis system in which a speaker being enrolled provides a plurality of utterances at least one of which is subjected to a comparison operation based on one or more others of the plurality of utterances, and in accordance with the results of the comparison the system decides whether to reject or accept a said utterance, the system storing characteristics of the speech input to it by the speaker derived from analysis only of non-rejected utterances.
Claim 3
Dependent Claim
43 words
A method according to claim 1 in which a first plurality of utterances provided by the speaker being enrolled are combined to provide a comparison reference, and one or more further utterances is compared with the comparison reference in a said comparison operation.
Claim 4
Dependent Claim
33 words
A method according to claim 2 in which, if a said further utterance is accepted in a said comparison operation, it is combined with the comparison reference to provide an updated comparison reference.
Claim 5
Dependent Claim
51 words
A method according to claim 2 or claim 3 in which, if a predetermined number of said further utterances are accepted before the number which are rejected reaches a rejection threshold, the system stores speech characteristics derived from analysis of the utterances of the first plurality and the accepted further utterances.
Claim 6
Independent Claim
80 words
A method according to any one of claims 2 to 4 in which, if the number of said further utterances which are rejected reaches a rejection threshold before a predetermined number are accepted, at least one of the utterances of the first plurality is rejected and replaced by a replacement utterance provided by the speaker being enrolled, and a new comparison reference is provided by combining any remaining utterances of the first plurality and the or each said replacement utterance.
Claim 7
Independent Claim
24 words
A method according to any one of claims 1 to 5, in which the durations of utterances are used in the said comparison operation.
Claim 8
Independent Claim
25 words
A method according to any one of claims 1 to 6, in which the cepstral co-efficients of utterances are used in the said comparison operation.
Claim 9
Independent Claim
62 words
A method of enrolling a speaker in a speech analysis system in which the system derives and stores characteristics of speech input to it by the speaker, the system deriving the said characteristics from utterances of a plurality of different texts input to it by the speaker and storing data representing the co-variability of the said characteristics between utterances of different texts.
Claim 10
Independent Claim
62 words
A method of enrolling a speaker in a speech analysis system in which the system derives and stores characteristics of speech input to it by the speaker, the system deriving the said characteristics from utterances of a plurality of different texts input to it by the speaker and storing data representing the co-variability of the said characteristics between utterances of different texts.
Claim 11
Dependent Claim
26 words
A method according to claim 8 in which respective characteristics specific to utterances by the speaker of respective texts are derived and stored by the system.
Claim 12
Dependent Claim
77 words
A method according to claim 9, in which at least one parameter of each said speech utterance is analysed, the said stored respective characteristics of the respective texts comprising a combination of the analysis results for a plurality of utterances of each text , and the said data represents or is derived from the co-variance between respective texts of the differences between the analysis results and the said combination of the analysis results for each respective text.
Claim 13
Independent Claim
48 words
A method of identifying a spoken utterance in a period of input waveform, in which the input waveform is analysed to discover the portion or portions thereof having pitch, and the spoken utterance is defined as comprising a portion or a plurality of closely spaced portions having pitch.
Claim 14
Independent Claim
48 words
A method of identifying a spoken utterance in a period of input waveform, in which the input waveform is analysed to discover the portion or portions thereof having pitch, and the spoken utterance is defined as comprising a portion or a plurality of closely spaced portions having pitch.
Claim 15
Dependent Claim
47 words
A method according to claim 11, in which the input waveform is also analysed with respect to a parameter other than pitch, and the spoken utterance is defined as comprising the said portion or portions together with any adjacent portions defined by means of the said parameter.
Claim 16
Dependent Claim
14 words
A method according to claim 12, in which the said parameter is instantaneous energy.
Claim 17
Dependent Claim
123 words
A method according to claim 12 or claim 13, in which a first instant is defined as the beginning of the portion, or the beginning of the first of the closely spaced portions, having pitch, a second instant is defined as the end of the portion, or the end of the last of the closely spaced portions, having pitch, a third instant is defined as the instant most closely preceding the first instant at which the said parameter has a predetermined value, a fourth instant is defined as the instant most closely following the second instant at which the said parameter has a predetermined value, and the spoken utterance is identified as beginning at the third instant and ending at the fourth instant.
Claim 18
Dependent Claim
50 words
A method according to claim 14, in which the predetermined value for the third instant and the predetermined value for the fourth instant are each defined in terms of a proportion of the maximum value of the parameter during the said portion or plurality of closely spaced portions having pitch.
Claim 19
Dependent Claim
30 words
A method according to claim 14 or claim 15, in which the predetermined value for the third instant and the predetermined value for the fourth instant are the same value.
Claim 20
Independent Claim
117 words
A method of recognising a speaker by analysing a test utterance obtained from the speaker and comparing the analysis results with data representing speech characteristics of a particular speaker, in which the test utterance is first analysed to identify the pitch periods of the voiced portions thereof, and the test utterance is subsequently analysed to determine the value of at least one parameter in each of a plurality of analysis periods of the test utterance, the analysis periods being selected so that each bears a similar timing relationship to a respective identified pitch period, an artificial identification of pitch periods being made in the first analysis in a preset manner during un-voiced portions of the test utterance.
Claim 21
Independent Claim
117 words
A method of recognising a speaker by analysing a test utterance obtained from the speaker and comparing the analysis results with data representing speech characteristics of a particular speaker, in which the test utterance is first analysed to identify the pitch periods of the voiced portions thereof, and the test utterance is subsequently analysed to determine the value of at least one parameter in each of a plurality of analysis periods of the test utterance, the analysis periods being selected so that each bears a similar timing relationship to a respective identified pitch period, an artificial identification of pitch periods being made in the first analysis in a preset manner during un-voiced portions of the test utterance.
Claim 22
Dependent Claim
21 words
A method according to claim 17, in which the said artificially identified pitch periods are shorter than the said analysis periods.
Claim 23
Dependent Claim
71 words
A method according to claim 17 or claim 18 in which the first analysis only identifies pitch periods falling within a permitted range, the longest period in the range being at least twice the shortest period, and for voiced portions of the test utterance having a voice pitch above the permitted range the first analysis identifies pitch periods which are an integer multiple of the periods of the said voice pitch.
Claim 24
Independent Claim
38 words
A method according to any one of claims 17 to 19, in which the first analysis only identifies pitch periods not exceeding a maximum length, and the length of the said analysis periods exceeds the said maximum length.
Claim 25
Independent Claim
65 words
A method of speaker recognition in which, in respect of each of a plurality of test utterances obtained from a speaker to be identified, speech characteristics obtained from the test utterance are compared with stored speech characteristics for a particular speaker, so as to obtain a corresponding plurality of comparison results, the said plurality of comparison results being used to derive a single recognition output.
Claim 26
Independent Claim
65 words
A method of speaker recognition in which, in respect of each of a plurality of test utterances obtained from a speaker to be identified, speech characteristics obtained from the test utterance are compared with stored speech characteristics for a particular speaker, so as to obtain a corresponding plurality of comparison results, the said plurality of comparison results being used to derive a single recognition output.
Claim 27
Dependent Claim
49 words
A method according to claim 21 in which the said comparison results comprise values representing respective degrees of similarity between each said test utterance and the stored speech characteristics with which it is compared, and the values are used in a statistical analysis to derive the single recognition output.
Claim 28
Dependent Claim
32 words
A method according to claim 21 or claim 22 , in which different said test utterances are compared with stored speech characteristics for speech by the said particular speaker of different texts.
Claim 29
Dependent Claim
46 words
A method according to claim 23 in which data relating to the co-variance between the said speech characteristics of the different texts is stored, and the said comparison results are used in combination with the said co-variance values in the derivation of the single recognition output.
Claim 30
Independent Claim
29 words
A method according to any one of claims 21 to 22 in which an F-test is used in the derivation of the single recognition output from the comparison results.
Claim 31
Independent Claim
122 words
A method of recognising a speaker by analysing a test utterance obtained from the speaker, comparing the analysis results with data representing speech characteristics of a particular speaker to obtain a value for the difference between them, and comparing the said value with a threshold, in which the threshold is determined by obtaining a first plurality of said values, for respective utterances known to have been provided by the said particular speaker, obtaining a second plurality of said values, for respective utterances known to have been provided by speakers of other than the said particular speaker, and selecting the threshold in dependence on at least one value of the first plurality of values and one value of the second plurality of values.
Claim 32
Independent Claim
122 words
A method of recognising a speaker by analysing a test utterance obtained from the speaker, comparing the analysis results with data representing speech characteristics of a particular speaker to obtain a value for the difference between them, and comparing the said value with a threshold, in which the threshold is determined by obtaining a first plurality of said values, for respective utterances known to have been provided by the said particular speaker, obtaining a second plurality of said values, for respective utterances known to have been provided by speakers of other than the said particular speaker, and selecting the threshold in dependence on at least one value of the first plurality of values and one value of the second plurality of values.
Claim 33
Dependent Claim
55 words
A method according to claim 26 in which the threshold is selected as the average of the value, amongst the first plurality of values, representing the greatest difference from the data representing speech characteristics of the particular speaker, and the value, amongst the second plurality of values, representing the smallest difference from the said data.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.