A novel signal subspace speech enhancement method

To obtain clean speech signal in strong noise environment is an important issue. We proposed a novel method in this article which combined with the signal subspace and auditory masking effect of human ear. With the relationship between the Eigen values of signal subspace and the frequency domain, the diagonal matrix of the gain function was calculated by using a variable Lagrange multiplier. The simulation waveform and spectrogram show that the proposed algorithm has good effect. Comparative Experiment also has a large signal-to-noise increase, it can balance well between reducing the noise and improving speech intelligibility; Informal audition results indicate the proposed algorithm can smooth the noise in the transition process between noise and speech.