What it is
VocaVoca makes a UTAU voicebank from your own voice: the app shows a prepared script, you just read or record what it asks for, and the voicebank forms automatically — cut, marked, and ready to use.
Making a UTAU voicebank the usual way is exhausting: finding a reclist yourself, recording and naming files one by one, then a long round of manual oto-ing. I once gave up on making one myself because of that. VocaVoca is built to cut that path short.
The principles it holds to:
- It still sounds like a voicebank. The identity of a voice synthesizer is kept. This is not AI-model voice cloning.
- Automatic gives a first result, manual perfects it. Every automatic result can be adjusted.
- Lightweight. The core has to run on an ordinary laptop, with no graphics card required.
- Everything in OpenUtau stays, including compatibility with UTAU voicebanks.
Built on OpenUtau
VocaVoca is a fork of OpenUtau, a vocal synthesis editor for the UTAU community by StAkira and its contributors, under the MIT license. That is why the stack follows OpenUtau: C#/.NET with Avalonia. The piano roll, the voice engine, and UTAU compatibility are not rewritten; what is added is the road from recording to voicebank.
OpenUtau’s LICENSE.txt and copyright notice must be included in full and
unchanged, in the repo and in the app. Updates from official OpenUtau are
pulled through an upstream remote, and the built-in update checker that
points to official OpenUtau releases is turned off so it doesn’t overwrite
VocaVoca.
Stage 1: proven
The first stage — the recorder and auto-oto — counts as done when reading the script produces a Japanese CV voicebank usable in OpenUtau. That was reached on 10 October 2026 with the test voicebank “Myu”, recorded from my own voice:
- Guided recorder: metronome and guide tones, latency calibration, and re-recording per line without disturbing the other lines.
- Auto-oto and “Create Voicebank”: 101 finished samples from 19 reclist lines; 10 flagged for review. Every value can still be tidied in the oto editor.
- A test project that follows the recorded pitch, can be rendered, and sounds like the recording.
- Compatible outside: the Myu voicebank loads and sings in official OpenUtau (0.1.573-beta), the test project opens there, and it sounds the same as in VocaVoca.
- A Windows x64 package that runs without installing the .NET SDK.
The limits, as they are: only a basic Japanese CV voicebank so far. Noise in the Myu session recordings is still −42 to −49 dB, while the target for a release is below −60 dB. Plosive consonants on connected notes are still not crisp enough. The stable OpenUtau 0.1.565 release cannot open the project file yet, although the voicebank itself still works there.
Next
CVVC and Stage 2 (auto tuning), worked on together:
- CVVC transition samples. A CV voicebank only has “consonant from silence” samples. A natural vowel-to-consonant transition on connected notes can only come from samples that actually record that transition — and this is required for an English pack.
- Stage 2: auto tuning. A voicebank from Stage 1 sings from notes drawn in the piano roll, and auto tuning gives a first pitch line that can still be adjusted by hand.
The order after that is already written, with no time estimates: voice profile variations, a text door (type a sentence and the voicebank speaks it), a cover door, and a real-time voice filter. Voicebanks made with VocaVoca are also planned to be the voice of Project Nor-4.