BiSinger: Bilingual Singing Voice Synthesis

Although Singing Voice Synthesis (SVS) has made great strides withText-to-Speech (TTS) techniques, multilingual singing voice modeling remainsrelatively unexplored. This paper presents BiSinger, a bilingual pop SVS systemfor English and Chinese Mandarin. Current systems require separate models perlanguage and cannot accurately represent both Chinese and English, hinderingcode-switch SVS. To address this gap, we design a shared representation betweenChinese and English singing voices, achieved by using the CMU dictionary withmapping rules. We fuse monolingual singing datasets with open-source singingvoice conversion techniques to generate bilingual singing voices while alsoexploring the potential use of bilingual speech data. Experiments affirm thatour language-independent representation and incorporation of related datasetsenable a single model with enhanced performance in English and code-switch SVSwhile maintaining Chinese song performance. Audio samples are available athttps://bisinger-svs.github.io.