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

2026-07-30

AI Summary

This work proposes a native score-driven, end-to-end singing voice synthesis method that aligns more closely with real-world music composition workflows. Unlike existing systems that rely on predefined or explicitly predicted duration information, the proposed approach takes as input lyrics, pitch, symbolic note durations, and tempo, and employs an interleaved lyric–note representation together with an autoregressive diffusion model to jointly generate acoustic features and implicitly determine output length in a latent acoustic space—eliminating the need for explicit duration prediction. Trained on 2,300 hours of data, the model outperforms the strongest baseline by 0.42 CMOS points in naturalness and demonstrates superior performance in intelligibility, melodic controllability, and speaker similarity, achieving seamless compatibility with practical music creation pipelines.

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.

Problem

• singing voice synthesis
• duration prediction
• composition workflow
• score-native
• acoustic guidance

Innovation

• score-native
• singing voice synthesis
• autoregressive diffusion model
• duration-free synthesis
• symbolic music representation