Changelog — VoiceCast (Native STT / Vosk)

Important: Pre-release only (Voice Chat required)

Native mode is experimental and ONLY works on the Hytale pre-release builds that include voice chat.
If you updated VoiceCast and it suddenly doesn’t work, please make sure your server is running the pre-release build shown below:

• Required server build: 2026.03.05-9fdc5985d

(If you’re on a stable/non-voice build, the native hook won’t be available.)

Added

• Native voice-to-RootInteraction pipeline (experimental)
  • Intercepts the in-game voice chat stream and maps recognized phrases to VoiceCast Castables → triggers RootInteractions.
• Offline STT via Vosk
  • Supports Vosk models and recognition without any external web service.
• Auto-download & install of Vosk models
  • Optional automatic model download from the official Vosk model hosting (direct links).
  • Handles ZIP layouts safely and “flattens” the extracted folder so am/ and conf/ end up in the correct modelPath.

Changed

• Default behavior targeting Native workflows
  • Recommended default config now prefers Native (and can keep Web disabled by default if desired).
• Native startup flow
  • The native service can start immediately with a NOOP recognizer, then swap in Vosk once the model is ready (downloaded/unzipped).

Notes / Known limitations

• Native mode depends on the pre-release voice pipeline and may break between pre-release updates.
• Vosk accuracy varies a lot by mic quality, background noise, and model choice (small vs large).
• Large models can be ~1GB; by design, the default flow prefers small models unless overridden via ModelUrl.

If you find a regression, please include:

• server build string (must match the pre-release above),
• your VoiceCast.json,
• and the relevant VoiceCast logs around model install / recognizer init.