摘要 |
PURPOSE:To provide the speech recognition device which provides excellent recognition by varying the feature vector' even if the feature vector of a phoneme for the speech recognition varies owing to a change of speakers or variation in voicing speed because of the articulation coupling of a continuous speech. CONSTITUTION:This speech recognition device is equipped with an analyzing means 2 which extracts a series of feature vectors from the speech, a phoneme series generating means 4 which generates a series of phonemes as a recognition reference, a context information extracting means 6 which extracts information on context from the outputs of the two means, a generation information storage means 7 stored with a series of coefficients, a standard pattern generating means 8 which obtains the standard pattern of the phoneme from the output of the context information extracting means 6 and the series of coefficients of the generation information storage means 7, probability density calculating means 101a-101d which input errors between the standard pattern and the feature vector of the input speech and find probability density functions, a mixing and adding means 105 which weights the obtained probability density functions by branch probability whose sum is restricted to 1 and adds the results, and a decision means 14 which finds the similarity from the output of the mixing and adding means 105 and decides the recognition result. |