Skip to main content

VRCT v3.5.0

· 2 min read

VRCT v3.5.0 is out — the first stable release since v3.4.3. It adds new translation and OBS integration features and rebuilds audio handling for stability.

Highlights​

Translation​

  • Default offline (CTranslate2) model changed to NLLB-200 distilled 600M.
  • New OpenAI-compatible endpoint as an independent translation provider — point it at any OpenAI-compatible base URL.
  • Switching engines now falls back unsupported languages to the default automatically; unsupported languages are greyed out in the picker.
  • DeepL is API-only now (the web-scraping engine was removed).

New features​

  • OBS browser source served by a built-in HTTP server — add one OBS browser source and show VRCT subtitles on your stream. See OBS Browser Source.
  • Follow the actively-playing audio endpoint: VRCT tracks peak levels and switches to the device that is actually producing sound.
  • Speech-recognition failures are now written to the system log.

Installer & telemetry​

  • The installer lets you pick the GPU or CPU edition and downloads only what is needed.
  • config.json is now preserved across upgrades.
  • Telemetry is limited to a once-per-day app_started event plus per-error-code events. No conversation content or personal data is ever sent.

Stability​

  • Audio processing was largely rebuilt: PyAudio calls are serialized under a single lock (no more native crashes), device switching is synchronous, stalled streams recover automatically, and the microphone open has an 8-second timeout.
  • OSCQuery setup gives up after a few retries instead of retrying forever; OSC itself keeps running.

The v3.4.3 documentation stays available from the version dropdown in the top navigation bar.