A Bilingual Speech Synthesis System of Standard Malay and Indonesian Based on HMM-DNN
Feng Chen, Jian Yang, Lixuan Zhao · 2020
This paper designs and implements a Standard Malay and Indonesian bilingual speech synthesis system based on HMM-DNN. Standard Malay and Indonesian have high homology and relatively few electronic language resources. We combine the corpus of these two similar languages and introduce speaker codes to examine bilingual speech synthesis system based on HMM-DNN, and compare it with the monolingual speech synthesis system. The methods used include: sharing synthesis units between Standard Malay and Indonesian, designing unified context attributes and question set, speaker adaptive training with speech corpus of these two languages, and synthesizing speech using speaker-dependent Standard Malay and Indonesian acoustic models. Experiments show that the quality of the Malay speech and Indonesian speech synthesized by the system is not only superior to the traditional Hidden Markov Model (HMM)-based bilingual speech synthesis system, but also superior to the HMM-DNN-based monolingual speech synthesis system.