EfficientSing: A Chinese Singing Voice Synthesis System Using Duration-Free Acoustic Model and HiFi-GAN Vocoder

Abstract

In this paper, we present EfficientSing, a Chinese singing voice synthesis (SVS) system based on a non-autoregressive duration-free acoustic model and HiFi-GAN neural vocoder. Different from many existing SVS methods, no auxiliary duration prediction module is needed in this work, since a newly proposed monotonic alignment modeling mechanism is adopted. More-over, we follow the non-autoregressive architecture of Effi-cientTTS with some singing-specific adaption, making training and inference fully parallel and efficient. HiFi-GAN vocoder is adopted to improve the voice quality of synthesized songs and inference efficiency. Both objective and subjective experimental results show that the proposed system can produce quite natu-ral and high-fidelity songs and outperform the Tacotron-based baseline in terms of pronunciation, pitch and rhythm.