Phase-locked Loop (PLL) Based Phase Estimation in Single Channel Speech Enhancement

Abstract

Conventional speech enhancement techniques are based on the modification of noisy spectral magnitude. In the reconstruction of the enhanced signal, noisy phase is combined with the modified noisy spectral magnitude. Recent studies on the importance of phase in enhancement process shows that the clean speech phase improves the quality of the enhanced signal. This work focused on Phase-Locked Loop (PLL) based time-domain approach for estimating the clean speech phase from noisy speech signal. The proposed technique is compared with the conventional approaches where noisy phase is used in the reconstruction of the enhanced signal. Here, Log-Likelihood Ratio (LLR), Weighted Spectral Slope (WSS) distance and Perceptual Evaluation of Speech Quality (PESQ) are used as performance measures. From experimental results, it is observed that the speech quality and intelligibility improved significantly with the proposed method over existing methods.