A study on recognition of continuous Chinese speech based on stochastic trajectory models

After pointed the unreality of the three basic HMM assumptions, this paper introduces the theory and the advantage of Stochastic Trajectory Models (STMs) which would possibly overcome these problems caused by HMM assumptions. STM represents the acoustic observations of a speech unit as clusters of trajectories in a parameter space. The trajectories are modeled by mixture of probability density functions of random sequence of states. After analyzing the characteristics of Chinese speech, the acoustic units for continuous Chinese speech recognition based on STM are discussed and phone-like units are suggested. The performance of continuous Chinese speech recognition based on STM is studied using VINICS system. The experimental results demonstrate the efficiency of STM and the consistency of phone-like units.