A Compression Algorithms for Hidden Markov Model-Based Speech Synthesis Systems

This paper proposes an efficient way of compressing parameters needed for hidden Markov model (HMM)-based statistical parametric speech synthesis (SPSS) systems. By analyzing the statistical characteristics of HMM model parameters, this paper proposes various quantization methods to minimize the footprint of HMM model parameters. Subjective and objective test results verify that the proposed algorithm not only reduces the memory size, but also maintains the perceptual quality of synthesized speech.