First International Conference on Spoken Language Processing (ICSLP 90)
The speech analysis method based on a glottal source model is important for efficiently constructing a rule-based speech synthesis system, since it is easy to control parameters, it can obtain smooth spectral trajectories and it can reduce memory capacity to store the parameters. This paper proposes a new speech analysis method based on the glottal source model. The proposed method uses glottal source and vocal tract inverse filterings in the frequency domain, and the model parameters are calculated so as to minimize inverse filtered error. The method can significantly reduce the computation amount and can assure vocal tract filter stability. Speech analysis synthesis experiments were carried out. The experimental results show that the speech spectrum is estimated more accurately and the estimated pole trajectory is smoother in the proposed method than in the conventional one. Further, synthetic speech is natural and intelligible. The results indicate that the proposed method is suitable for a rule-based speech synthesis system.
Bibliographic reference. Funaki, Keiichi / Mitome, Yukio (1990): "A speech analysis method based on a glottal source model", In ICSLP-1990, 45-48.