| 摘要 |
<p>A noise reduction system used for transmission and/or recognition of speech comprises a speech analyzer analyzing a noisy speech input signal thereby converting the speech signal into feature vectors (a set of feature quantities) such as autocorrelation coefficients, and a neural network receiving the feature vectors of the noisy speech signal as its input and extracting from a codebook, generated by previously clustering a set of feature vectors of a noise-free speech signal (a table storing prototype vectors corresponding to the feature vectors of the noise-free speech signal together with corresponding indices of the prototype vectors), the index of the prototype vectors corresponding to the noise-free equivalent of the noisy speech input signal, so that the feature vectors of speech are read out from the codebook on the basis of the index delivered as an output from the neural network, and the speech input can be reproduced on the basis of the feature vectors of speech read out from the codebook.</p> |
