SoulX-Singer: Towards High-Quality Zero-Shot Singing Voice Synthesis

2026-02-08

AI Summary

This work proposes the first industrial-grade, highly robust zero-shot singing voice synthesis (SVS) system, addressing the prevalent limitations of existing open-source SVS models—namely, poor robustness and inadequate zero-shot generalization in real-world deployment. Trained on over 42,000 hours of multilingual vocal data, the system leverages MIDI or melodic representations as conditioning inputs to enable controllable, cross-lingual, and multi-style singing synthesis. To facilitate reliable evaluation of zero-shot SVS, the authors also introduce SoulX-Singer-Eval, a rigorously disentangled benchmark specifically designed for this purpose. The proposed system achieves state-of-the-art synthesis quality in Mandarin, English, and Cantonese, while demonstrating exceptional zero-shot generalization across diverse musical contexts.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Computer Vision: Computational Photography, Image & Video Synthesis

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Systems and Infrastructure for Web, Mobile and WoT: Decentralized Web and Fediverse systems

Abstract

While recent years have witnessed rapid progress in speech synthesis, open-source singing voice synthesis (SVS) systems still face significant barriers to industrial deployment, particularly in terms of robustness and zero-shot generalization. In this report, we introduce SoulX-Singer, a high-quality open-source SVS system designed with practical deployment considerations in mind. SoulX-Singer supports controllable singing generation conditioned on either symbolic musical scores (MIDI) or melodic representations, enabling flexible and expressive control in real-world production workflows. Trained on more than 42,000 hours of vocal data, the system supports Mandarin Chinese, English, and Cantonese and consistently achieves state-of-the-art synthesis quality across languages under diverse musical conditions. Furthermore, to enable reliable evaluation of zero-shot SVS performance in practical scenarios, we construct SoulX-Singer-Eval, a dedicated benchmark with strict training-test disentanglement, facilitating systematic assessment in zero-shot settings.

Problem

• singing voice synthesis
• zero-shot
• robustness
• generalization
• industrial deployment

Innovation

• zero-shot singing voice synthesis
• controllable singing generation
• multi-lingual SVS
• MIDI-conditioned synthesis
• zero-shot evaluation benchmark