| 摘要 |
PROBLEM TO BE SOLVED: To provide a noise suppression method, a device and a program, capable of outputting emphasis voice with high quality, by attaining both the lower remaining noise in a non-voice section and lower output distortion, in a voice section, even when noise has large power with respect to voice. SOLUTION: In the present invention, an input signal is converted into a frequency domain signal; a first presumption noise is determined, by using the frequency domain signal, and a suppression coefficient is defined by using the first presumption noise and the frequency domain signal; a temporal output is determined, by weighting the frequency domain signal with the suppression coefficient; and second presumption noise is determined by using the temporary output. A section which is likely to be voice, and a section which is likely to be non-voice are defined by using the second presumption noise. A corrected suppression coefficient is determined by correcting the suppression coefficient, so that distortion may be lowered in the section which is likely to be voice, and that remaining noise is lowered in the section which is likely to be non-voice. Noise is suppressed by weighting the frequency domain signal with the corrected suppression coefficient. COPYRIGHT: (C)2010,JPO&INPIT |
