TuneLab 2.0 voice synthesis plugin for native NEUTRINO v3 CPU inference.
The plugin runs the original t.bin, p.bin, s.bin, and v.bin model
chain directly through ONNX Runtime. It does not call NEUTRINO.exe and does
not apply HNSEP or other audio post-processing.
- Windows x64
- TuneLab
release/2.0.0 - A NEUTRINO v3 installation containing at least one complete voice model
- .NET 8 SDK for building
Place the TuneLab source and this repository next to each other:
workspace/
|-- TuneLab/
\-- TuneLab-Neutrino/
Clone the matching TuneLab source with:
git clone --branch release/2.0.0 https://github.com/LiuYunPlayer/TuneLab.git.\build.ps1The installable package is written to:
artifacts/TuneLab.NeutrinoV3-win-x64.tlx
Open TuneLab's extension settings and add one or more voicebank directories.
The first row is blank by default, and the + button adds another directory.
Each entry may point to a complete NEUTRINO installation, its model directory,
or one individual voicebank directory such as model/ZUNDAMON.
A normal installation looks like:
NEUTRINO/
|-- model/
| \-- MERROW/
| |-- t.bin
| |-- p.bin
| |-- s.bin
| |-- v.bin
| \-- info.toml
Leaving the list blank keeps automatic discovery enabled. Overlapping entries are safe: each physical model directory is loaded only once. Phoneme conversion always uses the dictionary bundled with this plugin, so no dictionary is read from any configured voicebank directory.
Voice models are not included in this repository. Follow the licenses of NEUTRINO and each voice library.
- NEUTRINO v3 only
- CPU inference only
- Japanese kana, supported direct phonemes, and common romaji input
- TuneLab phoneme layout, pinned phonemes, pitch editing, rests, and
continuation lyrics such as
-,+, and+~ SHFCstyle shift automation from -1200 to +1200 cents with compensated F0
The plugin source code is available under the MIT License. NEUTRINO, TuneLab, ONNX Runtime, and voicebank files remain subject to their own licenses.