UtaiSynthesizer
utaisynthesizer.net
- Free plan
- Windows
What UtaiSynthesizer is
UtaiSynthesizer is a free, open-source Windows DAW that turns written scores into singing vocals with voice-conversion models. Its Score2ConVec model maps notation to features an SVC voice can sing without a human guide recording. RVC and So-VITS-SVC backends support shallow diffusion and multi-speaker blending. The workspace includes a piano roll, multi-track timeline, per-clip node workflow, vocal separation, training monitor and vocal-range extension. Dictionary-backed lyrics cover Chinese, English, Japanese, German, French, Spanish and Italian. The app imports USTX, UST, MIDI, audio and voice-model packages, and exports audio, UST, USTX and MIDI, including WAV, FLAC, MP3, OGG, OPUS and M4A. Training runs locally; inference does not require Python. Singer models are not included, so users import or train their own. It supports Windows 10 or 11 x64, is licensed under AGPL-3.0, and downloads are SHA-256 and minisign verified.
Who it suits
It suits Windows users creating synthetic vocals from notation who can supply or train voice models. It is also aimed at users wanting local processing and an open-source workflow.
Good to know, and what to watch
- Free and open source
- Score-to-singing without guide vocals
- Seven lyric languages
- Local model training
- Broad import and export support
- Windows 10 or 11 x64 only
- Singer models are not included
- AMD and Intel hardware support is limited or experimental
The verdict
UtaiSynthesizer provides a local, score-driven singing workflow with multiple SVC backends and extensive audio and notation formats. Users must bring their own singer models and check hardware support before training.
What the maker publishes
- DAW features
- Features include a piano roll, multi-track timeline, per-clip node workflow, vocal separation, training monitor, and vocal-range extension. source
- DAW features
- Features include a piano roll, multi-track timeline, node workflow, vocal separation, training monitor, and vocal-range tools. source
- Download integrity
- The security policy says downloads are SHA-256 verified and updates are minisign verified. source
- Downloads
- The official site links to a Windows download, and the repository specifies Windows 10 or 11 x64. source
- Import and export
- It exports audio, UST, USTX, and MIDI, and imports USTX, UST, MIDI, audio, and voice-model packages. source
- Import and export
- It exports audio, UST, USTX, and MIDI, and imports USTX, UST, MIDI, audio, and voice-model packages. source
- Integrity
- The repository says downloads are SHA-256 verified and auto-updates are minisign signed. source
- Languages
- The app supports dictionary-backed lyrics in seven languages: Chinese, English, Japanese, German, French, Spanish, and Italian. source
- Languages
- The app has a trilingual Chinese, English, and Japanese interface and supports lyric pronunciation in seven languages. source
- License
- The repository is licensed under AGPL-3.0. source
- License
- The project repository is licensed under AGPL-3.0. source
- Limitations
- The README says AMD runtime coverage is limited to gfx1103-class integrated GPUs and Intel XPU support is experimental and unverified on real hardware. source
- Local processing
- The maker describes voice-model training as on-device and says inference runs without requiring Python. source
- Models
- The repository and official installer do not include singer voice models; users import or train their own. source
- Offline operation
- The app runs locally and says it connects only for update checks, model and asset downloads, and an optional mirror connectivity test. source
- Platform
- The official site offers a Windows download, and the project README specifies Windows 10 or 11 x64. source
- Purpose
- UtaiSynthesizer is a singing-voice synthesis DAW that turns written scores into vocals using voice-conversion models. source
- Purpose
- UtaiSynthesizer is a singing-voice synthesis DAW that turns scores into vocals using voice-conversion models. source
- Recording and training
- The maker says a dozen minutes of dry vocals and about an hour or two of training can create a voice model. source
- Score-to-singing
- Its Score2ConVec model converts notation into features that let an SVC voice model sing without an intermediate human vocal recording. source
- Security reporting
- The project asks users to report vulnerabilities through GitHub's private vulnerability reporting rather than public issues. source
- Singing workflow
- Its Score2ConVec model maps a score into ContentVec features so an SVC model can sing from notation without a human guide vocal. source
- Support
- The maker directs bug reports and feature requests to GitHub Issues and lists QQ and Discord community groups. source
- Support
- The project directs bug reports and feature requests to GitHub Issues and links to QQ and Discord community groups. source
Checked against the maker's pages on 5 Oct 2026.