| 摘要 |
PROBLEM TO BE SOLVED: To make it possible to reproduce a more naturally synthesized speech. SOLUTION: Speech data are divided into two bands centering 1.2kHz through a band-pass filter group 8, and they are converted into LPC(Linear Prediction Coding) coefficients by LPC analyzers 41 and 42. A LPC inverse filter 5 outputs a residual error signal to speech output data from an A/D converter 2 by using the LPC coefficients of the two divided bands. The residual signals are divided into three bands partitioned by 1.5kHz and 2.5kHz through a band-pass filter group 6. A band-division residual signal is taken out as a sound source signal (electric power) for each divided band by sound analyzers 71, 72, and 73. Thus, voiced speech/unvoiced speech intermingled in each band are separated from each other. |
