Speech recognition apparatus
AssigneeKabushiki Kaisha Toshiba
InventorHiroshi Matsuura
Life of this patent
Estimating the term…
Abstract
A similarity memory stores a syllable recognition candidate and its similarity (j) supplied from a syllable recognition section at a specific storage position defined by reference syllable data and its syllable position (i), stores a subsidiary similarity (j-10) at an (i+1)th syllable position of the same syllable name, and stores a subsidiary similarity (j-20) at an (i+2)th syllable position of the same syllable name. The subsidiary similarity (j-10) is also stored at the (i+1)th syllable position of the same syllable name. A coincidence computing section computes a sum of similarities of a series of n syllables from the similarity memory and computes a coincidence.
Claims2
A speech recognition apparatus comprising:
- means for storing reference patterns corresponding to reference phonemes, respectively;
- means for extracting individual patterns of phonemes from an input utterance;
- means for comparing individual patterns of phonemes which have been extracted from an input utterance with the reference patterns and producing a comparison result;
- means for obtaining a positive similarity value j of at least one reference phonemes based on the comparison result;
- memory means having a plurality of areas corresponding to all the reference phonemes, phoneme positions being assigned in time-axis direction for each of the reference phonemes and the areas being provided corresponding to the respective phoneme positions and storing the similarity values;
- means for storing in the area defined by the reference phoneme and the phoneme position, the similarity value j obtained by said similarity value obtaining means;
- means for obtaining subsidiary similarity values having smaller values than the similarity value j;
- means for temporarily storing the subsidiary similarity values in areas defined by immediately before and after the phoneme position of the same reference phoneme;
- means for preliminarily storing words to be recognized;
- means for extracting the similarity value j from said similarity value storing means and the subsidiary similarity values from said similarity value temporary storing means in accordance with the individual phonemes of a word from said word storing means;
- means for accumulating the extracted similarity values from said similarity extracting means;
- and
- means for recognizing the word based on the accumulated result from said accumulating means.
A speech recognition apparatus comprising;
- means for storing reference patterns corresponding to reference phonemes, respectively;
- means for extracting individual patterns of syllables from an input utterance;
- means for comparing individual patterns of syllables which have been extracted from an input utterance with the reference patterns and producing a comparison result;
- means for obtaining a positive integer similarity value j of at least one reference syllable based on the comparison result;
- memory means having a plurality of areas corresponding to all the reference syllables, syllable positions being assigned in time-axis direction for each of the reference syllables and the areas being provided corresponding to the respective syllable positions and storing the similarity values;
- means for storing, in the area defined by the reference syllable and the syllable position, the similarity value j obtained by said similarity value obtaining means;
- means for obtaining subsidiary similarity values having smaller values than the similarity value j;
- means for temporarily storing the subsidiary similarity values in areas defined by immediately before and after the syllable position of the same reference syllable;
- means for preliminary storing words to be recognized;
- means for extracting the similarity value j from said similarity value storing means and the subsidiary similarity values from said similarity value temporary storing means in accordance with the individual phonemes of a word from said word storing means;
- means for accumulating the extracted similarity values from said similarity extracting means;
- and
- means for recognizing the word based on the accumulated result from said accumulating means.
Dependencies are inferred from the claim text.