| 摘要 |
PROBLEM TO BE SOLVED: To compress an input voice waveform of a friction sound with high compressibility and to make it free from sense of incongruity. SOLUTION: A LPC coefficient is obtained based on correlation of input voice waveforms discriminated as voiceless sounds (s1), while the power of input voice waveform for each section previously set is obtained (s2), data of these input voice waveforms is converted to data of these LPC coefficient and power, and it is delivered to a voice synthesizing side. In the voice synthesizing side, numerical values of the prescribed range decided based on a sampling period used by a system are generated with random numbers in a process in which linear prediction using the LPC coefficient is performed (s23). These random numbers are made pitches of impulses, while numerical values of the prescribed range decided based on the maximum value of amplitude of an input voice waveform are generated with random numbers, these random numbers are made amplitude of the impulse (s24) thereby emphasizing correlation of a voice waveform by controlling amplitude of an impulse response waveform in obtained some timing. |
