A Wavelet Packet Based Approach for Speech Enhancement Using Modulation Channel Selection

Abstract

In this paper, a wavelet packet based speech enhancement system is proposed for noise reduction. In the proposed method, a modulation channel selection is used as a thresholding function for de-noising. Three levels 8 sub-band wavelet packet decomposition is used and all sub-bands are given to threshold function for noise suppression. This novel modulation channel selection is based on calculation of true signal-to-noise ratio (SNR) by thresholding with local SNR of -7 dB. The presented method is used for noise suppression in single-channel speech patterns. Objective and subjective parameters are used for performance evaluation of this method. The performance of the proposed method is also compared with spectral subtraction, mband, mmse, test-psc, idbm, klt, and pklt. The proposed method give maximum intelligibility and quality in compared to other given methods. MATLAB 7.14 is used for simulation.

A Novel Underwater Acoustic Signal Denoising Algorithm for Gaussian/Non-Gaussian Impulsive Noise

“…Sumithra et al [30] proposed a trimmed thresholding method to achieve a compromise between the soft and the hard thresholding in order to effectively enhance the quality of the SOI in the background noise. Singh et al [31] used the modulation channel selection as the threshold function and proposed a denoising method based on wavelet packet that was able to obtain a higher output SNR compared with other algorithms. Based on the improved thresholding function, Zhang et al [32] proposed a dual-tree complex wavelet transform (ITF-DTCWT) denoising algorithm to ensure the approximate translation invariance and reduce the signal distortion.…”

Realce E Reconhecimento De Voz Contínua Em Ambientes Adversos

“…Em [107] o método de realce baseado em waveletdenosing é usado para atenuar mais o sinal após aplicar uma função de ganho para minimizar o erro médio quadrático, a fim de separar quadros de voz e não voz com base na probabilidade calculada pela razão de verossimilhança de dois modelos de mistura de Gaussianas (GMMs). Em [108] uma seleção de canal de modulação é usada como função de limiar para denoising. Nessa proposta se faz uma decomposição de três níveis obtendo 8 sub-bandas usando a função wavelet mãe symlet4 onde as sub-bandas de alta frequência são passadas pela função de limiar para supressão de ruído.…”