| 摘要 |
<p>PURPOSE:To obtain a synthesized voice of high quality by sampling the logarithmic power spectrum obtained from a voice waveform by a fundamental frequency and applying a cosine series model to an obtained sample point to obtain a spectrum envelope and obtaining a mel cepstrum coefficient from the spectrum envelope. CONSTITUTION:An analyzing part 1 obtains a short-time power spectrum from short-time voice waveform data and samples this spectrum in a position of integer-fold fundamental frequency and applies a cosine series model to the obtained sample point to obtain the spectrum envelope. A parameter converting part 2 calculates the mel cepstrum coefficient from this spectrum envelope. The mel cepstrum coefficient obtained by the parameter converting part is inputted to a synthesizing part 3 as the filter coefficient of a mel logarithmic spectrum approximating filter, and this part 3 produces a synthesized voice. Thus, the quality of the synthesized voice is improved.</p> |
