Log-Domain Speech Feature Enhancement Using Sequential MAP Noise Estimation and a Phase-sensitive Model of the Acoustic Environment

摘要: In this paper we present an MMSE (minimum mean square error) speech feature enhancement algorithm, capitalizing on a new probabilistic, nonlinear environment model that effectively incorporates the phase relationship between clean and corrupting noise in acoustic distortion. The estimator based phase-sensitive is derived it achieves high efficiency by exploiting single-point Taylor series expansion to approximate joint probability of noisy as multivariate Gaussian. As integral component also sequential MAP-based nonstationary estimator. Experimental results Aurora2 task demonstrate importance corruption process captured reported performs significantly better than phase-insensitive spectral subtraction (54% error rate reduction), noticeably our previous state-of-the-art technique [2] (7% under otherwise identical experimental conditions recognition.