speech synthesis

VocalRender: Score-Native Singing Voice Synthesis for Real-World Composition

arXiv:2607.27768

summary

The paper introduces VocalRender, a system that can directly synthesize singing voices from musical scores—including lyrics, pitches, note values, and tempo—without needing separate duration predictions, using an autoregressive diffusion model to generate audio.

Abstract

Existing singing voice synthesis systems often require predefined durations, explicit duration prediction, or time-aligned acoustic guidance, which limits their compatibility with practical composition workflows. We propose VocalRender, a score-native system that directly synthesizes singing from lyrics, pitches, symbolic note values, and tempo. It uses an interleaved lyric--note representation and an autoregressive diffusion model to generate continuous acoustic latents while predicting the output length, eliminating the need for explicit duration prediction. Trained on a 2,300-hour singing dataset, VocalRender achieves strong intelligibility, strong melody control, and high speaker similarity across both in-domain and out-of-domain benchmarks. Notably, it outperforms the strongest baseline by $0.42$ points in naturalness CMOS, demonstrating the effectiveness of our proposed score-native architecture.

Topics & keywords

#singing voice synthesis #score-native synthesis #diffusion models #duration-free generation #music compositionautoregressive diffusion modelinterleaved lyric-note representationacoustic latent generationnaturalness CMOSlarge singing dataset