| 摘要 |
PURPOSE:To synthesize voice with high quality by providing a spectrum parameter storage memory, phoneme environment label storage memory, code conversion part, spectrum parameter time sequence generation part, continuity scale calculation part and spectrum parameter connection control part. CONSTITUTION:Character code trains to be synthesized are converted to the sequence of phoneme environment labels by a code conversion part 18 and while referring to a phoneme environment label storage memory 12, the phoneme environment label sequence reads out the spectrum parameter of the phoneme having the coincident phoneme environment label from a spectrum parameter storage memory 11. By connecting those parameters, the spectrum parameter time sequence is generated by a spectrum parameter time sequence generation part 22. Next, a connectivity scale calculation part 23 calculates connecting distortion showing the quality of connection at each connecting part of the spectrum parameter time sequence and further, a spectrum parameter connection control part 21 decides the combination of phoneme to miniaturize the sum of the connecting distortion in the entire synthesized voice. Thus, the voice can be synthesized with high quality. |
