Robust Speech Enhancement Using Dabauchies Wavelet Based Adaptive Wavelet Thresholding for the Development of Robust Automatic Speech Recognition: A Comprehensive Review

Abstract

Developing a robust Automatic Speech Recognition (ASR) system is a major challenge in speech signal processing research. These systems perform exceedingly well in clean environments. However, the performance of these systems is not acceptable when the spoken signal is corrupted by several environmental and other artificial noises. The efficiency of any ASR system depends on several factors such as size of the vocabulary, native language influences, transmission channel, emotional and health state of the speaker, age of the speaker, designed speech corpus, size of the dataset, training and testing strategy and its preprocessing and other challenges. It is well known fact that the presence of noise in speech signal degrades its perceptual quality and intelligibility and hence ASR system performance is also affected. So, in this paper Dabauchies Wavelet based time adaptive Bayes thresholding algorithm is proposed with a custom Wavelet Packet Decomposition and Reconstruction Tree. The proposed system performance is evaluated on the Private Kannada Dataset and TIMIT dataset. The results reveal the effectiveness of the proposed system in various SNR levels such as − 10, − 5, 0, 5, 10, 15, 20, 25 and 30 dB. The article begins with introductory insights on ASR, Physiological process of speech production and perception in Humans, ASR jorgans, the architecture of ASR, and barriers associated with the ASR design. The work also focus on dataset design, baseline speech enhancement methods. This work provides comprehensive review to Wavelet based speech enhancement approach to the research scholars pursuing research in the area of speech signal processing.