Voxxwire v1.3.0 — Hallucination Filtering, Smarter Translation & Settings Persistence

🎉 What's New in v1.3.0
🖥️ UI Fixes

Fixed dropdown misalignment in Speech Settings panel (Closes #2)

🧠 Whisper Hallucination Filtering

RMS energy gate filters out silence/noise before it reaches ASR
Built-in blocklist catches common phantom outputs ("thank you", "subscribe", etc.)
Repetition detection discards looping hallucinations

🌐 Smarter Translation

Non-English sources now use Whisper's native translate task for direct English output — much more accurate than Argos for hi/ja → en
Argos only handles en → target leg (its strongest direction)

⚡ Pipeline Improvements

Per-channel thread pools — mic and loopback no longer starve each other for CPU
Utterance merging — queued audio batches automatically when ASR falls behind
Retuned VAD threshold (0.2 → 0.35) and max speech duration (15s → 8s) for fewer false triggers and faster processing

💾 Settings Persistence

Device, language, and model selections now saved to user_settings.json and auto-restored on next launch

🧹 Housekeeping

Removed committed pycache/ files, added pywebview==6.1 dependency

📦 Upgrade: pip install -r requirements.txt
💻 Requires Windows 10/11 64-bit | 🔒 Fully offline | 🌍 49+ languages
🐛 Report a bug · 🌐 voxxwire.com