Singing Voice Synthesis

Singing voice synthesis studies how to generate and modify vocals while controlling lyrics, melody, singer identity, and expressive style. Common approaches include diffusion and flow-matching models, autoregressive audio generation, and neural vocoders, often with separate controls for linguistic content, pitch, and timbre. Systems are evaluated for intelligibility, melody and lyric adherence, naturalness, singer similarity, and performance across genres; related work also explores lyric editing, singing-style conversion, and unified speech-and-singing generation. A central challenge is making controls reliable across musical styles without sacrificing vocal identity, expression, or audio quality, especially when training data and genre-specific benchmarks are limited.