Dysarthria Voice Enhancement Using Novel WPT-TDO Algorithm

The dysarthic speech recognition (DSR) is very critical in human machine interaction system because of its lower speech intelligibility. The performance of the DSR is hugely depends upon the quality of speech samples. However, dysarthric speech is often having noise and artifacts which degrades the quality of voice. This paper presents the novel wavelet packet transform (WPT) based soft thresholding method for speech enhancement to minimize noise and artifacts present in the speech signal. The Tasmanian devil optimization algorithm (TDOA) is utilized for the hyper-parameter tuning of the WPT such as a number of decomposition level, wavelet filter type, and value of K. The effectiveness of the suggested WPT-TDO based speech enhancement is evaluated on the public UASpeech dataset based on peak signal to noise ratio (PSNR), mean absolute error (MAE) and mean square error (MSE).