Deep Feed-forward Sequential Memory Networks for Speech Synthesis
This work addresses runtime efficiency for speech synthesis applications, but it is incremental as it adapts an existing method from speech recognition to a new domain.
The paper tackled the high model complexity and inference cost of Bidirectional LSTM (BLSTM) in speech synthesis by applying Deep Feed-forward Sequential Memory Networks (DFSMN), resulting in comparable speech quality while drastically reducing model complexity and generation time.
The Bidirectional LSTM (BLSTM) RNN based speech synthesis system is among the best parametric Text-to-Speech (TTS) systems in terms of the naturalness of generated speech, especially the naturalness in prosody. However, the model complexity and inference cost of BLSTM prevents its usage in many runtime applications. Meanwhile, Deep Feed-forward Sequential Memory Networks (DFSMN) has shown its consistent out-performance over BLSTM in both word error rate (WER) and the runtime computation cost in speech recognition tasks. Since speech synthesis also requires to model long-term dependencies compared to speech recognition, in this paper, we investigate the Deep-FSMN (DFSMN) in speech synthesis. Both objective and subjective experiments show that, compared with BLSTM TTS method, the DFSMN system can generate synthesized speech with comparable speech quality while drastically reduce model complexity and speech generation time.