Cascaded speech enhancement technique in highly nonstationary noise environment

This paper discusses on the enhancement of noisy speech in highly nonstationary noise environment. The concept of Minimum Mean Square Error (MMSE) inspired Data driven noise power (DDNP) estimation and geometrical approach to Spectral Subtraction (GASS) is used in cascade in order to obtain a better Segmental Signal to Noise Ratio (SSNR) and good Perceptual Evaluation of Speech Quality (PESQ) score. This proposed method gives better results for low SNR speech signals compared to iterative enhancement of the individual GASS and DDNP algorithms.