| 摘要 |
<P>PROBLEM TO BE SOLVED: To attain improvement of sound quality in selectively reducing a noise signal by estimating the noise signal from an input signal. <P>SOLUTION: A noise suppression gain calculation section 32 calculates a noise suppression gain in each of bands from a post-SNR and a pre-SNR. A pre-SNR calculation section 31 calculates the pre-SNR in each of bands using post-SNRs of a preceding frame and a current frame, a noise suppression gain of the preceding frame and a weight coefficient α(k, b). The weight coefficient α(k, b) in each of bands calculated by an α calculation section 30 is adaptively varied in accordance with a signal state. Namely, the weight coefficient α(k, b) is updated to approach a maximum value αMAX(b) in a band (b) determined as a noise, and immediately updated to a minimum value αMIN(b) in a band (b) determined as a non-noise. Thus, the accuracy (follow-up property) of the noise suppression gain in each band calculated by a noise suppression gain generation section 15 can be improved and improvement of sound quality can be attained. <P>COPYRIGHT: (C)2012,JPO&INPIT |
