Performance analysis of adaptive wavelet denosing by speech discrimination and thresholding

The performance of automatic speech recognition can be elevated either at the front end by pattern recognition, speech enhancement or by a powerful classifier at the backend. A adaptive time space strategy is applied to Bayes threshold and Universal threshold denoising techniques, to enhance the signal to noise ratio of noisy speech and principle of speech discrimination from silence is employed before speech enhancement, to prevent the over thresholding of speech coefficients for maintaining the intelligibility of speech. The experimental results reveal that performance of time space adaptive Bayes shrink denoising principle outperforms the later in enhancing signal to noise ratio of noisy speech while retaining the speech intelligibility.