HMM-Based Speech Recognition System with Preprocessing Enhancement Module

In my graduation project, I dedicated myself to developing a complete Hidden Markov Model (HMM)-based speech recognition system. This project represents not only my substantial effort but also demonstrates my technical capabilities and passion for this field. The system implementation involved several key components: MFCC (Mel-Frequency Cepstral Coefficients) feature extraction for converting speech signals into numerical representations, Baum-Welch algorithm for HMM parameter training, and Viterbi algorithm for decoding the most probable word sequences. Furthermore, I integrated an enhancement module prior to the recognition system to ensure improved accuracy and stability. This preprocessing module employed various digital signal processing techniques including spectral subtraction for noise reduction, voice activity detection (VAD) using energy threshold algorithms, and signal normalization for amplitude consistency. The implementation required sophisticated frame-based processing with overlapping windows to maintain temporal continuity. Throughout the project development, I gained deep understanding of speech processing technologies and HMM statistical modeling approaches. This comprehensive learning experience has thoroughly prepared me to tackle future challenges in speech technology and pattern recognition applications.