On a modified cepstral pitch control technique for the high quality text-to-speech type system

摘要

In the area of speech synthesis, the waveform coding methods are mainly used to maintain intelligibility and naturalness of synthetic speech. However, it is difficult to apply the waveform coding methods to the synthesis by rule since this methods does not separate both the excitation information and vocal tract information from a speech signal. This paper proposes a modified pitch alteration method that can reduce the spectrum distortion by reconstructing the pitch altered speech signal using both the formant component in the frequency domain and the phase component in the time domain. This has only a little spectrum distortion of 1.18% for 50% pitch change.