Speech recognition system with an adaptive acoustic model
At the time of the speaker adaptation, first feature vector generation sections ( 7, 8, 9 ) generate a feature vector series [C i, M ] from which the additive noise and multiplicative noise are removed. A second feature vector generation section ( 12 ) generates a feature vector series [S i, M ] including the features of the additive noise and multiplicative noise. A path search section ( 10 ) conducts a path search by comparing the feature vector series [C i, m ] to the standard vector [a n, m ] of the standard voice HMM ( 300 ). When the speaker adaptation section ( 11 ) conducts correlation operation on an average feature vector [S^ n, m ] of the standard vector [a n, m ] corresponding to the path search result Dv and the feature vector series [S i, m ], the adaptive vector [x n, m ] is generated. The adaptive vector [x n, m ] updates the feature vector of the speaker adaptive acoustic model ( 400 ) used for the speech recognition.
1. A speech recognition system comprising:
a standard acoustic model having a standard vector generated according to information on speech;
a first feature vector generation section for reducing noise from an input signal generated from uttered speech corresponding to a designated text to generate a first feature vector;
a second feature vector generation section for generating a second feature vector from the input signal having the noise; and
a preparation section for generating an adaptive vector based on the first feature vector, the second feature vector and the standard vector, and preparing a speaker adaptive acoustic model suitable for the uttered speech.
2. The speech recognition system according to claim 1 , wherein the preparation section compares the first feature vector with the standard vector to obtain a path search result; and
the preparation section coordinates the second feature vector with the standard vector according to the path search result to generate the adaptive vector.
3. The speech recognition system according to claim 1 , wherein the noise includes additive noise and multiplicative noise.
4. The speech recognition system according to claim 3 , wherein the first feature vector generation section includes an additive noise reduction section for reducing the additive noise from the input signal to generate an additive-noise reduced signal.
5. The speech recognition system according to claim 4 , wherein the additive noise reduction section applies a transformation to the input signal to generate a first spectrum and subtracting an additive noise spectrum corresponding to the additive noise from the first spectrum.
6. The speech recognition system according to claim 4 , wherein the first feature vector generation section includes a cepstrum calculator for applying cepstrum calculation to the additive-noise reduced signal.
7. The speech recognition system according to claim 6 , wherein the first feature vector generation section includes a multiplicative noise reduction section for reducing the multiplicative noise by subtracting the multiplicative noise from the first feature vector.
8. The speech recognition system according to claim 7 , wherein the first feature vector contains a plurality of time-series first feature vectors; and
the multiplicative noise reduction section calculates a time average of the time-series first feature vectors for estimating the multiplicative noise.
9. The speech recognition system according to claim 1 , wherein the second feature vector generation section applies at least cepstrum calculation to the input signal to generate the second feature vector.