Speech Recognition: Converting Speech Signals into Text Using Machine Processing

This article explores speech recognition technology, which enables machines to convert speech signals into corresponding text or commands through identification and comprehension processes. Our research employs the DTW (Dynamic Time Warping) algorithm for preliminary investigation into isolated word recognition. We successfully implemented isolated word speech recognition in the MATLAB environment, incorporating practical code implementations for feature extraction and template matching. The implementation involves key steps such as preprocessing audio signals, extracting MFCC (Mel-Frequency Cepstral Coefficients) features, and applying DTW for time-series alignment between input patterns and reference templates. Our analysis summarizes DTW's primary characteristics, including its effectiveness in handling temporal variations, and its limitations concerning computational complexity and scalability issues. Through this article, readers will gain deeper insights into speech recognition technology and the practical application of DTW algorithm in this field, with specific references to MATLAB coding approaches and algorithmic optimization techniques.