SPEECH ADAPTATION DEVICE, WORD SPEECH RECOGNIZING DEVICE, CONTINUOUS SPEECH RECOGNIZING DEVICE AND WORD SPOTTING DEVICE

摘要

PURPOSE: To provide a speech adaptation device capable of executing adaptation with high accuracy even if the utterance contents of input speeches are unknown. CONSTITUTION: This speech adaptation device has an analysis section 1 which converts the input speeches to the time series of characteristic vectors, vocabulary non-dependent standard patterns 3 which have connection relations of such acoustic units having >=1 categories, having >=1 acoustic units in the respective categories and capable of receiving the chains of any acoustic units appearing in the speeches, a prematching section 2 which makes the time series of the characteristic vectors of the input speeches obtd. from the analysis section 1 and the vocabulary non-dependent standard patterns 3 correspondent to each other and determines the average vector of every category of the input speeches and the vocabulary non-dependent standard patterns 3 from the correspondent parts of every category of the characteristic vectors of the input speeches and the vocabulary non-dependent standard patterns 3 and an adaptation section 4 which subjects either or both of the time series of the characteristic vectors of the input speeches or the vocabulary non-dependent standard patterns to correction by using the average vector of the every category calculated by this prematching section 2.