Method and apparatus for preparing acoustic model and computer program for preparing acoustic model

A method for preparing an acoustic model used for speech recognition, comprising the steps of:

• (a) selecting speech data acoustically similar to voice data of a user of the speech recognition from a plurality of speech data of a plurality of speakers;
• (b) superimposing noise in an environment in which the speech recognition is used on the speech data selected in the step (a);
• (c) preparing sufficient statistics using noise-superimposed speech data obtained in the step (b);
• (d) preparing an,acoustic model using the sufficient statistics prepared in the step (c).

An apparatus for preparing an acoustic model used for speech recognition, comprising:

• a storage section for storing sufficient statistics prepared using speech data in each of a plurality of groups, the plurality of groups being obtained by grouping noise-superimposed speech data according to acoustic similarity;
• a first selection section for selecting a group acoustically similar to voice data of a user of the speech recognition from the plurality of groups;
• a second selection section for selecting sufficient statistics acoustically similar to the voice data of the user from the sufficient statistics in the group selected by the first selection section;
• and
• a model preparation section for preparing an acoustic model using the sufficient statistics selected by the second selection section.

An apparatus for preparing an acoustic model used for speech recognition, comprising:

• a storage section for storing a plurality of speech data of a plurality of speakers;
• a selection section for selecting speech data acoustically similar to voice data of a user of the speech recognition from the speech data stored in the storage section;
• a noise superimposing section for superimposing noise in an environment in which the speech recognition is used on the speech data selected by the selection section;
• a sufficient statistic preparation section for preparing sufficient statistics using noise-superimposed speech data obtained by the noise superimposing section;
• and
• a model preparation section for preparing an acoustic model using the sufficient statistics prepared by the sufficient statistic preparation section.

A computer program for preparing an acoustic model used for speech recognition, the program tailoring a computer to function as the means of:

• (a) storing sufficient statistics prepared using speech data in each of a plurality of groups, the plurality of groups being obtained by grouping noise-superimposed speech data according to acoustic similarity;
• (b) selecting a group acoustically similar to voice data of a user of the speech recognition from the plurality of groups;
• (c) selecting sufficient statistics acoustically similar to the voice data of the user from the sufficient statistics in the group selected by the means (b);
• and
• (d) preparing an acoustic model using the sufficient statistics selected by the means (c).

A computer program for preparing an acoustic model used for speech recognition, the program tailoring a computer to function as the means of:

• (a) storing a plurality of speech data of a plurality of speakers;
• (b) selecting speech data acoustically similar to voice data of a user of the speech recognition from the speech data stored in the means (a);
• (c) superimposing noise in an environment in which the speech recognition is used on the speech data selected by the means (b);
• (d) preparing sufficient statistics using noise-superimposed speech data obtained by the means (c);
• and
• (e) preparing an acoustic model using the sufficient statistics prepared by the means (d).