Abstract: To solve the problem of oversimplified speech short-time spectral amplitude distribution model in speech signal processing, an estimation method of speech amplitude spectrum based on hidden Markov model is proposed. This estimation method uses Rayleigh mixture model as speech amplitude spectrum distribution and employs hidden Markov model to divide speech signal into different states. In each state, there is accordingly a group of Rayleigh mixture model parameters. By the division of speech signal into different states, this method can achieve speech classification and build more accurate model for speech signal short term spectrum.

Key words: speech signal processing, Rayleigh mixture model, hidden Markov model