Single channel speech enhancement using a complex spectrum method

Speech enhancement plays an important role in speech communication systems. Speech signal enhancement in an additive noise environment in speech recognition and speaker verification system is still a challenging task. In speech enhancement process, the spectral analysis method has more advantageous than other methods due to its simplicity and effective localization of noise components in the signal. But, this method does not analyze the phase information for efficient speech enhancement. This present work proposes a modified phase spectrum compensation method for speech enhancement in a single channel environment that analyzes both the magnitude and phase spectrum of the speech signal. The performance of the proposed method is compared with that of three conventional methods (Spectral Subtraction (SSUB), Minimum Mean Square Error (MMSE) estimator, and Phase Spectrum Compensation (PSC)) through four different objective measures: Log spectral distance (LSD), log likelihood ratio (LLR), itakura-siato (IS) measure, and short-time objective intelligibility (STOI). The experimental results show that the three objective measures (LLR, LSD, and STOI) of the proposed method gives better results over the conventional methods in four different noise Signal to Noise Ratio (SNR). © 2020 Elsevier B.V., All rights reserved.