Simple VRC Translator

⚠️ This project is coded with AI. I am determined to squash any bugs and am happy to take any feedback in issues.

Windows desktop app for translating speech or typed text into VRChat's OSC chatbox.

Features

• Translate typed text into up to three target languages.
• Send original text, translated text, or both through VRChat OSC.
• Speech recognition with WebView speech or local CPU Whisper.
• Download, select, and delete local Whisper models from the app.
• Choose how many CPU threads Local Whisper uses.
• Optional English curse-word masking for Local Whisper.
• Auto-send speech, clean punctuation, bracket or number translations, and choose startup listening.
• Test OSC, inspect diagnostics, and open OSC logs from Advanced settings.
• Built-in update checks from GitHub releases.

Install

Download SimpleVRCTranslator.exe from the latest release, then run it. No installer or administrator access needed.

VRChat OSC must be enabled. Default app settings send to 127.0.0.1:9000.

Speech engines

WebView speech uses Windows WebView2 speech recognition. It needs an internet connection and does not allow the app to control provider-side profanity filtering. If WebView2 is missing, install Microsoft Edge WebView2 Runtime.

The app keeps microphone capture open and buffers phrases in memory, including up to 750 ms before speech detection. After 350 ms of silence (or a ten-second limit), it feeds the saved mono audio to a fresh WebView recognition session at normal speed. This adds replay latency but preserves audio through recognition startup and retries. A phrase is retried at most twice; a 45-second queue limit reports an error instead of silently losing queued speech. Stop Listening clears buffered audio and releases the microphone. Audio is not saved to disk. Extreme Debugging logs capture levels, phrase timing, retries, and recognition text. Buffered input requires WebView support for SpeechRecognition.start(audioTrack).

Local Whisper (CPU) records from your default microphone and runs the selected downloaded GGML model locally. Recognition works without network access after model download. Translation still uses Google Translate. Start with Tiny; larger models use more memory and take longer to process speech.

Privacy and updates

WebView speech is handled by its provider. Typed text and recognized speech sent for each selected target language go to Google Translate. Local Whisper recognition stays on your computer after model download. This app does not collect telemetry.

At startup, app checks stable releases in this repository. Update downloads are checked against the release SHA-256 before replacement and restart.

Build from source

Requirements:

• Windows
• Node.js with pnpm
• Rust stable toolchain
• LLVM with libclang.dll available through LIBCLANG_PATH

pnpm install
$env:LIBCLANG_PATH = 'C:\Program Files\LLVM\bin'
pnpm build:desktop

Output: src-tauri\target\release\SimpleVRCTranslator.exe.

Report problems

Open an issue with your Windows version, app version, selected speech engine, and steps that reproduce the problem. Do not include private chat text in logs or screenshots.

License

Source-visible project. No license is granted.