An application of the Bayesian time series model and statistical system analysis for F0 control

F0 (fundamental frequency) control of human speech reduction is studied by using both a stochastic time series model and a system analysis with a vector autoregressive (VAR) model. We use two-dimensional time series data of F0 (ff0 and sf0) obtained through the transformed auditory feedback (TAF) experimentation system developed by one of the authors, sf0 is extracted from speech data that includes prolonged phonation of the vowel id, and the signal ff0 is extracted from frequency-modulated feedback speech by white Gaussian noise. Most of the data had mean-nonstationary characteristics. The stochastic procedure decomposes the mean-nonstationary and other components in one stage without pre-manufacturing the data. The cyclical components around the mean-nonstationary components are assumed to be generated by the VAR model. We can execute a stochastic system analysis using the estimates of the VAR model to analyze the physical characteristics of the data. We also performed a simulation study using the estimates obtained by the model to discover the role of F0 control under the situation in which we completely lost our hearing ability. The results clearly indicate the dynamic properties of F0 control for each segment which occur with each breath taken during a sustained tone. (C) 1998 Elsevier Science B.V. All rights reserved.