Short-Term Analysis of Speech Signals

This MATLAB implementation performs short-term analysis of speech signals, featuring voiced/unvoiced classification and pitch period estimation. The analysis employs frame-based processing where the speech signal is divided into short overlapping segments (typically 20-30ms frames) using windowing functions like Hamming or Hanning windows. For voiced/unvoiced decision, the algorithm calculates short-term energy and zero-crossing rate features - voiced segments show higher energy with lower zero-crossing rates, while unvoiced segments exhibit lower energy with higher zero-crossing rates. Pitch period estimation utilizes time-domain methods such as autocorrelation analysis or cepstral analysis to identify the fundamental frequency. This analysis provides crucial insights into speech signal characteristics and variations, thereby enhancing the performance and accuracy of speech recognition and synthesis applications. By examining short-term speech properties, we can deeply explore sound formation principles and propagation patterns, establishing a foundation for advanced research and applications in speech processing.