Efficient Acoustic Model with Enhanced Feature Extraction Technique for Automatic Tamil Speech Recognition

Abstract

The process of converting an acoustic speech signal into a human readable newlinetext format is known as automatic speech recognition (ASR). ASR systems have newlineprogressed beyond recognizing across isolated digits to spontaneous speech, opening newlineup a wide range of useful applications in several industries. The Gaussian Mixture newlineModel - Hidden Markov Model (GMM-HMM) strategy is used by conventional newlinesystems to take use of the acoustic component of ASR. One of the most challenging newlineissues affecting ASR is environmental unpredictability. Any signal distortion, such as newlinevariations in speech output and other environmental factors, decreases the accuracy of newlinean ASR. The feature extraction and acoustic model make up the two primary newlinecomponents of the speech recognition system. By removing the appropriate speech newlinecharacteristics, a reliable and accurate speech recognition system may be created. The newlineadoption of inadequate feature extraction techniques is one of the key reasons for the newlinesignificant failure of any speech recognition system. The relevant feature selection newlineproblem is attempted to be solved in this thesis. Deep learning models used by ASR newlinesystems for acoustic modeling have generated a lot of research attention newline