SoulX-Singer is a singing voice synthesis model that generates voices for singers it has not been tuned on. It offers two ways to guide a performance: melody-based F0 contours or MIDI score notes, with control over pitch, rhythm, and expression. Its conversion component can transform raw singing audio into a target singer’s voice while retaining the melody, rhythm, and lyrics, without requiring lyric or MIDI transcripts. The system supports Mandarin Chinese, English, and Cantonese, along with timbre cloning, cross-language synthesis, and lyric edits that preserve natural prosody. Tools cover vocal separation, dereverberation, pitch extraction, voice activity detection, and lyric and note transcription. Users can edit MIDI lyrics, phoneme alignment, pitches, and durations before synthesis. A generation and conversion demo and a separate MIDI editor are available through Hugging Face Spaces; the repository also supports local inference. Code and weights are under Apache 2.0, and the project is free. Maintainers caution that automatic preprocessing can misalign singing with lyrics and notes, so manual correction may be needed.
Who it is for
SoulX-Singer suits researchers and developers seeking a free, open-source tool for singing synthesis or voice conversion. Its MIDI controls and local deployment options may also suit users who want to edit and run their own synthesis workflow.
What is good
- Free code and weights under Apache 2.0
- Supports Mandarin Chinese, English, and Cantonese
- Controls pitch, rhythm, and expression through F0 or MIDI
- Includes singer-timbre cloning and voice conversion
- Offers web demos and local inference
What to know first
- Automatic preprocessing may misalign lyrics and notes
- Local inference requires setup with Conda, Python 3.10, and pip dependencies
- Users are asked to avoid unauthorized impersonation and deceptive audio
Verdict
SoulX-Singer combines voice synthesis, conversion, and MIDI editing with web demos and a local workflow. Check and correct automatic preprocessing results before relying on them, and use voices with appropriate consent.
SoulX-Singer plans and pricing
All plansCompared on AI song cover generators
- Voice cloning
- Yes
- Vocal input
- Yes
- MIDI support
- Yes
- Stem export
- Yes
- Supported languages
- 3 languages
- Export formats
- MIDI
- Commercial use
- allowed
