No abstract

Waveform generator is a key component in voice conversion. Recently, WaveNet waveform generator conditioned on the Mel-cepstrum (Mcep) has shown better quality over standard vocoder. In this paper, an enhanced WaveNet model based on spectrogram is proposed to further improve voice conversion performance. Here, Mel-frequency spectrogram is converted from source speaker to target speaker using an LSTM-RNN based frame-to-frame feature mapping. To evaluate the performance, the proposed approach is compared to an Mcep based LSTM-RNN voice conversion system. Both STRAIGHT vocoder and Mcep-based WaveNet vocoder are elected to produce the converted speech for Mcep conversion system. The fundamental frequency (F0) of the converted speech in different systems is analyzed. The naturalness, similarity and intelligibility are evaluated in subjective measures. Results show that the spectrogram based WaveNet waveform generator can achieve better voice conversion quality compared to traditional WaveNet approaches. The Mel-spectrogram based voice conversion can achieve significant improvement in speaker similarity and inherent F0 conversion.

Speech duration is an important component in statistical parameter speech synthesis(SPSS). In LSTM-RNN based SPSS system, the speech duration affects the quality of synthesized speech in two aspects, the prosody of speech and the position features in acoustic model. This paper investigated the effects of duration in LSTM-RNN based SPSS system. The performance of the acoustic models with position features at different levels are compared. Also, duration models with different network architectures are presented. A method to utilize the priori knowledge that the sum of state duration of a phoneme should be equal to the phone duration is proposed and proved to have better performance in both state duration and phone duration modeling. The result shows that acoustic model with state-level position features has better performance in acoustic modeling (especially in voice/unvoice classification), which means statelevel duration model still has its advantage and the duration models with the priori knowledge can result in better speech quality.