Auto-Regressive Analysis and Simulation of Speech Signal

This paper focuses mainly on auto-regressive analysis and simulation on speech signal as nonstationary time series. Simulating speech signal here means generating speech signal as nonstationary time series. Preprocessing of the known speech signal includes framing, windowing and endpoint detection and so on. This phrase mainly realizes the ascension of signal, detecting start and end points of signal. Preprocessing speech signal is the premise and basis of simulation, which plays an important role in speech recognition or simulation of speech signal. We analyze preprocessed speech signal as nonstationary time series with full order and time-varying order time-varying auto-regressive (TVPAR) models. This paper focuses on the way to analyze and simulate speech signal by TVPAR models. With parameters of the variance of error term and regressive coefficients, speech signal is simulated. Experimental results verified the TVPAR model and showed that the simulation method is practicable and the simulated speech signal likes the actual one.