Releases: raj20023/voxxwire
Release list
Voxxwire v1.4.0 — Real WASAPI Loopback, Live Config Reload & Transcript Export
Highlights
- Fixed mic/remote audio cross-contamination — loopback capture now uses real WASAPI loopback (via pyaudiowpatch) instead of a "Stereo Mix"-style input device, so "You" and "Remote" streams can no longer bleed into each other.
- Fixed loopback stalling during silence — a silent keep-alive stream keeps the Windows audio engine awake so the Remote channel never stops capturing during quiet periods.
- Fixed dropped mic audio ("input overflow") — reduced Whisper's CPU thread usage to leave headroom for the real-time audio callback, plus more robust capture buffering settings.
- Fixed a stale translation cache bug — a newly downloaded language pack now works immediately instead of requiring a full app restart.
- Save Configuration & Restart and Apply — change languages/devices/model without restarting the whole app.
- Download Transcript — export the full session (both channels, timestamps, original + translated text) to a text file.
- Clearer engine loading UI — a proper loading/error overlay with a Retry button, and per-channel "Listening / Failed / Disabled" status badges so it's obvious whether capture actually started.
- Hardened engine start/stop/restart against race conditions from rapid clicking.
Install
Download Voxxwire-Setup.exe below and run it.
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
Voxxwire v1.2.0 — Split-Screen UI, Auto Audio Detection & Improved Pipeline
🎉 What's New in v1.2.0
🖥️ Redesigned App UI
- Split-screen translation view — Left panel shows "🎙️ You" (your mic input + translation), right panel shows "🔊 Remote" (system audio + translation)
- Auto audio device detection — Automatically selects the best microphone and system audio devices (prefers headsets/headphones)
- Minimal device controls — Clean chip-style display with an option to manually change if needed
- BETA badge with link to report issues
🎙️ Better Voice Detection
- 2.5x microphone gain boost — Voice detected at normal speaking distance (no need to shout into the mic)
- Lowered VAD threshold (0.4 → 0.2) for improved sensitivity
- Faster response — Reduced minimum speech duration and silence detection for snappier translations
🔧 Robust Audio Pipeline
- No more dropped audio — VAD and ASR/Translation now run independently using a decoupled utterance queue
- Audio is never lost while a previous sentence is being processed
🌐 Website Updates
- Live download counter showing total downloads
- SmartScreen warning instructions for first-time installers
- Bug reporting link to GitHub Issues
📦 Download: Voxxwire-Setup.exe (211 MB)
💻 Requires: Windows 10/11 64-bit
🔒 Fully offline — No internet required after installation
🌍 49+ languages supported
⚠️ Note: Windows may show a SmartScreen warning on first run. Click "More info" → "Run anyway". We've submitted the app to Microsoft for whitelisting — this warning will be removed soon.
🐛 Found a bug? Report it here
🌐 Website: voxxwire.com
Voxxwire v1.1.0 — Better mic sensitivity + no audio dropped
🎙️ What's new in v1.1.0:
- Fixed: Audio no longer dropped while translating previous sentence
- Fixed: Voice now detected at normal speaking distance
- Improved: 2.5x mic gain boost for better sensitivity
- Improved: Faster response after pausing speech