Automatic Speech Recognition and Translation for Low Resource Languages

18Automatic Speech Recognition Based on Improved Deep Learning

Kingston Pal Thamburaj* and Kartheges Ponniah

Faculty of Languages and Communications, Sultan Idris Education University, Perak, Malaysia

Abstract

Tamil, along with other South Asian languages, consumes moderately petite responsiveness in the arena of natural language processing (NLP), specifically in automatic speech recognition (ASR). While ASR has seen significant advancements in English, European, and East Asian languages, there has been a lack of beneficial research in South Asian languages, including Tamil. In this chapter, a method for creating an ASR system for the Tamil language is presented. In order to effectively identify speech signals, this research effort offers an intelligent ASR system employing an improved deep learning (IDL) model. The least mean squares (LMS) filtering is used to remove noise initially. The features from noise-removed signals are then derived by mel-frequency cepstral coefficient (MFCC) technique to enhance the outcomes. Subsequently, a refined deep learning model based on recurrent neural networks (RNN) is employed for voice signal classification. Modified social spider optimization (MSSO) is developed. Sphinx4, an open-source speech recognition framework that makes use of enhanced deep learning to construct ASR systems, serves as the foundation for the suggested system. The chapter also aims to demonstrate the functionality of a speaker-independent ASR system tailored for ...