A New Improved Algorithm of Speech Enhancement Based on MCRA and Noncausal a Priori SNR Estimator

摘要

In view of the Minimum Mean Square Error of Log-Spectral Amplitude estimator (MMSE-LSA) noise power spectrum estimation algorithm does not adapt to the actual non-stationary noise and the accuracy of the estimation of prior SNR is not high on the traditional logarithmic domain. However, most of the nonstationary noises are encountered in real world. This paper puts forward a new improved speech enhancement algorithm of MMSE-LSA. Fist, it uses Minima Controlled Recursive Averaging(MCRA) to estimate noise which can track nonstationary effectively. Then combined with non-causal prior SNR estimation, it has higher accuracy of estimation of actual speech power spectrum, eliminating noise further. The simulation results show that under the non-stationary noise, the effect of restrain background noise and music noise residual has large improvement by the improved algorithm, the signal-to-noise ratio and the voice quality have been improved. In particular it is clear that the proposed algorithm has an improved capability to retain low amplitude voiced speech components in low SNR conditions.