Speaker verification

A method of speaker verification, comprising:

• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification;
• the method further comprising:
• forming the adjusted standard deviation from the standard deviation of the plurality of cohort scores and a prior tuning factor;
• wherein the prior tuning factor is obtained by, for each of a plurality of test utterances:
• comparing the test utterance against a second plurality of models of speech obtained from a second plurality of other speakers respectively;
• obtaining a second plurality of cohort scores from comparing the test input against the second plurality of models of speech;
• and
• calculating a standard deviation of the second plurality of cohort scores;
• and
• forming an average of said standard deviations calculated for each of the plurality of test utterances,
• wherein said average of said standard deviations is taken as the prior tuning factor.

A method of speaker verification, comprising:

• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification;
• the method further comprising:
• forming the adjusted standard deviation from the standard deviation of the plurality of cohort scores and a prior tuning factor;
• wherein the prior tuning factor is obtained by:
• measuring a property of the test input;
• and
• selecting the prior tuning factor based on the measured property.

A speaker verification system, comprising:

• an input, for receiving a test input representing speech;
• and a processor, wherein the processor is configured for:
• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification;
• wherein the processor is further configured for calculating the adjusted standard deviation σ2 from the standard deviation of the plurality of cohort scores and a prior tuning factor, as:
• σ22=γσ02+(1−γ)σ2
• where:
• σ is the standard deviation of the plurality of cohort scores,
• σn is the prior tuning factor.

A tangible non-transitory computer-readable medium, comprising code, for causing a processor to perform a method comprising:

• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification;
• the method further comprising:
• calculating the adjusted standard deviation σ2 from the standard deviation of the plurality of cohort scores and a prior tuning factor, as:
• σ22=γσ02+(1−γ)σ2
• where:
• σ is the standard deviation of the plurality of cohort scores,
• σn is the prior tuning factor.

A speaker verification system, comprising:

• an input, for receiving a test input representing speech;
• and a processor, wherein the processor is configured for:
• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification,
• the processor being further configured for:
• forming the adjusted standard deviation from the standard deviation of the plurality of cohort scores and a prior tuning factor,
• wherein the prior tuning factor is obtained by, for each of a plurality of test utterances:
• comparing the test utterance against a second plurality of models of speech obtained from a second plurality of other speakers respectively;
• obtaining a second plurality of cohort scores from comparing the test input against the second plurality of models of speech;
• and
• calculating a standard deviation of the second plurality of cohort scores;
• and
• forming an average of said standard deviations calculated for each of the plurality of test utterances,
• wherein said average of said standard deviations is taken as the prior tuning factor.

A tangible non-transitory computer-readable medium, comprising code, for causing a processor to perform a method comprising:

• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification,
• the method further comprising:
• forming the adjusted standard deviation from the standard deviation of the plurality of cohort scores and a prior tuning factor,
• wherein the prior tuning factor is obtained by, for each of a plurality of test utterances:
• comparing the test utterance against a second plurality of models of speech obtained from a second plurality of other speakers respectively;
• obtaining a second plurality of cohort scores from comparing the test input against the second plurality of models of speech;
• and
• calculating a standard deviation of the second plurality of cohort scores;
• and
• forming an average of said standard deviations calculated for each of the plurality of test utterances,
• wherein said average of said standard deviations is taken as the prior tuning factor.

A speaker verification system, comprising:

• an input, for receiving a test input representing speech;
• and a processor, wherein the processor is configured for:
• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification,
• the processor being further configured for:
• forming the adjusted standard deviation from the standard deviation of the plurality of cohort scores and a prior tuning factor,
• wherein the prior tuning factor is obtained by:
• measuring a property of the test input;
• and
• selecting the prior tuning factor based on the measured property.

A tangible non-transitory computer-readable medium, comprising code, for causing a processor to perform a method comprising:

• comparing a test input against a model of a user's speech obtained during a process of enrolling the user;
• obtaining a first score from comparing the test input against the model of the user's speech;
• comparing the test input against a first plurality of models of speech obtained from a first plurality of other speakers respectively;
• obtaining a plurality of cohort scores from comparing the test input against the plurality of models of speech obtained from a plurality of other speakers;
• obtaining statistics describing the plurality of cohort scores, wherein the statistics describing the plurality of cohort scores comprise a standard deviation;
• modifying said statistics to obtain adjusted statistics, wherein the adjusted statistics comprise an adjusted standard deviation;
• normalising the first score using the adjusted statistics to obtain a normalised score;
• and
• using the normalised score for speaker verification,
• the method further comprising:
• forming the adjusted standard deviation from the standard deviation of the plurality of cohort scores and a prior tuning factor,
• wherein the prior tuning factor is obtained by:
• measuring a property of the test input;
• and
• selecting the prior tuning factor based on the measured property.