Third International Conference on Spoken Language Processing (ICSLP 94)
This paper presents a new speech analysis method for waveform segment concatenation synthesis or PSOLA ( Pitch Synchronous Over-Lap and Add) synthesis. In the proposed method, two techniques are employed: a new algorithm for calculating the complex cepstrum and a technique for extracting waveform segments from human speech. Although the complex cepstrum analysis can estimate not only spectral envelope but also phase characteristics, every usual algorithm has its own problem in the application to voiced speech analysis. The proposed algorithm can solve these problems, and the combination of this technique with a residual excitation technique can extract waveform segments as accurate approximation of original human speech. Evaluation tests for the proposed method was carried out, i.e. analysis-synthesis experiment and pitch modification synthesis. In the subjective listening tests for both synthesized speech and pitch modified speech, the listeners could not recognize the distortions. This shows the effectiveness of the proposed method.
Bibliographic reference. Mitome, Yukio (1994): "Speech analysis technique for PSOLA synthesis based on complex cepstrum analysis and residual excitation", In ICSLP-1994, 1075-1078.