Speech Enhancement Based on Simple CASA Model

Based on the principle of computational auditory scene analysis (CASA) and the frequency masking effect of human ears, a new simple single channel CASA speech enhancement algorithm was proposed to reduce the auditory effect of non speech noises in the mixed inputs. The useful speech time frequency components can be picked up successfully by the proposed algorithm, and they are used to reconstruct the enhanced outputs. Lots of computer simulations show that the proposed CASA speech enhancement algorithm can improve the output SNR (about 10 dB) and the understandable ability of the outputs distinctly in both case of using the car noise and the white noise.