A Novel Wavelet Packet Speech Enhancement Algorithm Based On Time-Frequency Threshold

Threshold choosing is critical in wavelet speech enhancement. In the paper, a novel wavelet packet speech enhancement algorithm is presented based on time-frequency (TF) threshold. Different from the conventional methods in threshold choosing, e.g. invariant threshold and time-variant threshold, the proposed threshold is modulated according to speech TF details other than rough envelops adopted in the recent algorithms based on eager energy operator (TEO) and adaptive noise estimation (ANE). In the new algorithm, the speech TF information is obtained from the frequency-based pre-estimate, and the threshold is modulated with TF characteristic of the pre-estimate. Then via thresholding the wavelet packet coefficients, the contaminated speech can be denoised adaptively. Compared with the former wavelet based algorithms, the proposed algorithm offers more pleasant enhanced speech with less distortion and residual noise in additive Gaussian noise case. Experimental results show its better performance in subjective test, input and output SNR test and modified bark spectral distortion measurement (MBSD).