Time-Varying Modeling for Speech Signals. Part 1. Mathematical Modeling of Time-Varying Speech

摘要

Recent research on speech processing indicates that a rigorous study of time-varying models for speech analysis is urgently needed. In the report, a simple procedure is provided which shows that, under a regularity condition, virtually all speech models proposed in the literature are solutions of a corresponding, unique, time-dependent, difference equation. Consideration of some properties of the underlying physical production process also yields a time-dependent difference equation. A closed-form solution for a general, time-varying difference equation representative of the speech signal is derived and called the general time-varying model. To derive this solution, the concept of time-dependent poles, is applied. GTVM is a useful tool for providing possible acoustic-invariance cues for speech recognition.