Noise compensation in speaker-adaptive systems

Apparatus for adapting an acoustic model relating acoustic units to speech vectors, wherein said acoustic model comprises a set of acoustic model parameters related to a given speech factor and which enable the acoustic model to output speech vectors with different values of the speech factor,

• the system comprising:
• a receiver for receiving a sample of speech with a first value of the speech factor;
• and
• a processor configured to:
• (a) determine values of the set of acoustic model parameters which enable the acoustic model to output speech with said first value of the speech factor;
• and
• (b) employ said determined values of the set of acoustic model parameters in said acoustic model,
• wherein said sample of speech is corrupted by noise, and
• wherein said determination of the values of the set of acoustic model parameters comprises:
• (i) obtaining noise characterization parameters characterising the noise with which the sample of speech is corrupted;
• (ii) performing an acoustic model parameter generation algorithm on the sample of speech, thereby generating corrupted values of the set of acoustic model parameters;
• using the noise characterisation parameters to map said corrupted values of the set of acoustic model parameters to clean values of the set of acoustic model parameters, wherein the clean values of the set of acoustic model parameters are estimates of the acoustic model parameters which would be obtained by performing the acoustic model parameter generation algorithm on the sample of speech if the sample of speech were not corrupted by the noise.

A tangible, non-transitory carrier medium storing, in non-transitory form, computer readable code configured when performed by a processor of a computer to cause the processor to adapt an acoustic model relating acoustic units to speech vectors, wherein said acoustic model comprises a set of acoustic model parameters related to a given speech factor and which enable the acoustic model to output speech vectors with different values of the speech factor,

• the computer readable code being operative to cause the processor to:
• input a sample of speech with a first value of the speech factor;
• determine values of the set of acoustic model parameters which enable the acoustic model to output speech with said first value of the speech factor;
• and
• employ said determined values of the set of acoustic model parameters in said acoustic model,
• wherein said sample of speech is corrupted by noise, and
• wherein said determination of the values of the set of acoustic model parameters comprises:
• (i) obtaining noise characterization parameters characterising the noise;
• performing an acoustic model parameter generation algorithm on the sample of speech, thereby generating corrupted values of the set of acoustic model parameters;
• (iii) using the noise characterization parameters to map said corrupted values of the set of acoustic model parameters to clean values of the set of acoustic model parameters, wherein the clean values of the set of acoustic model parameters are estimates of the acoustic model parameters which would be obtained by performing the acoustic model parameter generation algorithm on the sample of speech if the sample of speech were not corrupted by the noise;
• and
• (iv) employing said clean values of the set of acoustic model parameters as said determined values of the set of acoustic model parameters.