Multiplicative update of AR gains in codebook-driven speech enhancement

Abstract

This paper presents a novel technique for estimating autoregressive (AR) parameters of speech and noise in the codebook-driven Wiener filtering speech enhancement method. We only train the shape codebook of speech spectrum offline, and the shape of noise spectrum is estimated online for solving the problem of noise classification. Unlike conventional codebook-driven methods, we exploit a multiplicative update rule to estimate the AR gains of speech and noise more accurately. Meanwhile, the Bayesian parameter-estimator without the noise codebook is also developed. Moreover, we achieve the goal of removing the residual noise between the harmonics of noisy speech by utilizing a very simple method, i.e., combining the codebook-driven Wiener filter with the speech-presence probability (SPP). The test results confirm the superiority of our method.