Method and apparatus for CELP coding an audio signal while distinguishing speech periods and non-speech periods

For the CELP (Code Excited Linear Prediction) coding of an input audio signal (S), an autocorrelation matrix (R), a speech/noise decision signal (v) and a vocal tract prediction coefficient (a) are fed to an adjusting section (111). In response, the adjusting section (222) computes a new autocorrelation matrix (Ra) based on the combination of the autocorrelation matrix of the current frame and that of a past period determined to be noise. The new autocorrelation matrix (Ra) is fed to an LPC (Linear Prediction Coding) analyzing section (103). The analyzing section computes a vocal tract prediction coefficient (a) based on the autocorrelation matrix (R) and delivers it to a prediction gain computing section (112). At the same time, in response to the above new autocorrelation matrix (Ra), the analyzing section (103) computes an optimal vocal tract prediction coefficient (aa) by correcting the vocal tract prediction coefficient (a). The optimal vocal tract prediction coefficient (aa) is fed to a synthesis filter (104).