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

Researchers have released SoulX-Singer, an open-source AI system for generating high-quality singing voices. It can create singing from a musical score or melody without needing prior training on a specific voice, a capability known as zero-shot synthesis. Trained on over 42,000 hours of vocal data, it supports Mandarin, English, and Cantonese. The team also released a dedicated benchmark to reliably test such systems in real-world, zero-shot scenarios.