Soulx Singer

SoulX-Singer - Zero-Shot Singing Voice Synthesis

High-fidelity zero-shot singing voice synthesis by Soul AI Lab (Shanghai Jiao Tong University). Generate realistic singing voices for any unseen singer.

When to Use

• User wants to generate singing voices from any reference
• User wants to clone a singer's voice
• User needs melody-conditioned singing synthesis
• User wants MIDI-based singing control
• User needs singing voice conversion (SVC)

Key Features

• Zero-shot singing voice synthesis (no fine-tuning needed)
• Melody (F0) and Score (MIDI) conditioning
• Timbre cloning across languages and styles
• Singing voice editing (modify lyrics keeping prosody)
• Cross-lingual synthesis
• Audio-to-Audio SVC (no transcription needed)
• 42,000+ hours of aligned vocals, lyrics, notes
• Languages: Mandarin, English, Cantonese

Installation

git clone https://github.com/Soul-AILab/SoulX-Singer.git
cd SoulX-Singer
pip install -r requirements.txt

Quick Start

Download Models

# SVS and SVC models
hf download Soul-AILab/SoulX-Singer --local-dir pretrained_models/SoulX-Singer
# Preprocessing models
hf download Soul-AILab/SoulX-Singer-Preprocess --local-dir pretrained_models/SoulX-Singer-Preprocess

Generate Singing Voice

from soulx_singer import SoulXSingerv1

model = SoulXSingerv1.from_pretrained("./pretrained_models/SoulX-Singer")

# Zero-shot singing from lyrics and reference voice
audio = model.synthesize(
    lyrics="Your lyrics here",
    melody_reference="reference_singing.wav",
    voice_reference="target_singer.wav"
)
audio.save("output.wav")

Singing Voice Conversion (SVC)

# Convert source singing to target voice
audio = model.svc(
    source_audio="source_singing.wav",
    target_reference="target_voice.wav",
    midi_transcribe=False  # Audio-to-audio, no transcription needed
)
audio.save("converted.wav")

Parameters

Related

• rvc: Voice conversion and singing synthesis
• heartmula: Open source music generation
• ycing-music-singer: Another singing voice synthesis model
• tcsinger2: Multilingual singing voice synthesis