PURPOSE:To obtain a better result of spectral approximation in both sound source and a voice meatus by taking continuity as a time series into consideration with regard to voice meatus characteristic, and thereafter, deriving again the sound source in accordance with necessity. CONSTITUTION:By being accompanied with a vocal cord/voice meatus analysis executed independently at every analytic section, first of all, whether a result obtained successively by an analyzing means 2 is unsuitable or not, and from a result of analysis of a section which precedes the section concerned, a voice meatus parameter value in the section concerned is prediction. A result of prediction and the present analysis result are compared, and in the case its difference is outside of a set range, the result of prediction is adopted as a result of analysis of a more appropriate voice meatus characteristic. In this case, in the case first selection of the sound source is unappropriate, in order to eliminate a newly determined voice meatus characteristic from its original sound signal, a voice meatus reverse filter 6 in which a result of this prediction is a coefficient is constituted, a sound signal of the section concerned is inputted from an input buffer 1 to this filter 6, and from an analysis result output means 9. In such a way, spectral approximate is improved in both the source and the voice meatus.
展开▼