MONAURAL NOISE SUPPRESSION BASED ON COMPUTATIONAL AUDITORY SCENE ANALYSIS

摘要

The present invention provides a sophisticated noise reduction system which can reduce the noise and echo components of the sound signal at the same time while limiting the level of the speech distortion. Acoustic signal is received and converted to the cochlear domain subband signals. The characteristics and the pitch may be identified and tracked in the sub-band signals. Then, the initial speech and noise models can be estimated from the probability analysis based on at least in part, the track pitch source. The speech and noise models can be resolved from the initial noise model and the speech and the noise reduction can be performed on a subband signal may be an acoustic signal from the noise reduced sub-band signal reconstruction.