Speech Recognition: Developing Algorithm for Speech Segmentation

This project investigates the development of an algorithm for the automatic detection of voiced and unvoiced segments in speech signals, crucial for improving the accuracy of speech recognition systems. The project begins with an introduction highlighting the limitations of current input devices and the advantages of speech recognition. It then outlines the research objectives, including a comprehensive literature review of existing methods, selection of appropriate windowing techniques, feature extraction, and the development of a noise-robust algorithm. The thesis is structured into seven chapters, starting with an introduction and a literature review that examines various approaches to speech classification, including energy-based methods, zero-crossing rate, auto-correlation, and neural networks. The project then delves into the fundamentals of speech signals and pre-processing techniques, feature extraction methods, and the algorithms used for classification. The results are presented through graphs and analysis, discussing the accuracy of the developed algorithm. The project aims to address the challenges in speech recognition, particularly in noisy environments, and proposes a generic optimal classification algorithm to enhance the accuracy of voiced and unvoiced speech segment classification. The project also includes an overview of speech signal processing and the applications of speech recognition across various fields, emphasizing the importance of speech segmentation for applications like speech synthesis, speech recognition, and speaker recognition.