| 摘要 |
PURPOSE: To efficiently encode a voice by utilizing not only prediction based on past encoded vectors but also correlations of adjacent frequencies in a current vector in encoding the voice. CONSTITUTION: A DFT(digital Fourier transform) means 22 subjects each frame of an input voice to DFT at a point TN in accordance with a pitch period TN to obtain a power spectrum, and a subtraction means 23 obtains the error between the power spectrum and its predicted value. An addition means 24 adds the error value and the predicted value of the power spectrum to obtain a reproduced power spectrum, and a predicted value operation means 25 adds an inter-frame predicted value obtained from the reproduced power spectrum and the preceding power spectrum and an intra-frame predicted value obtained from the power spectrum in the current frame to obtain a predicted value of the power spectrum. The voice is efficiently encoded by the pitch period TN and the error signal which are obtained in this manner. |
