Table of Contents
1. Quick Start & Installation
Voxxwire is designed to deliver sub-second, 100% offline speech-to-speech translation directly on Windows 10 and Windows 11 64-bit systems. Follow these steps to get started:
- Download the standalone installer (`Voxxwire-Setup.exe`) from the Download Page or Releases Changelog.
- Run the setup executable. If Windows SmartScreen displays a warning, click "More info" and select "Run anyway".
- Launch Voxxwire from your Desktop shortcut or Start Menu.
- On first launch, Voxxwire will automatically scan your system's input and output audio hardware.
2. Audio Device Setup (Microphone & System Loopback)
Voxxwire features a dual-channel audio architecture designed specifically for online meetings (Zoom, Google Meet, Microsoft Teams, Discord, or Skype):
🎙️ Microphone Channel (You)
Captures your spoken voice through your physical headset or microphone. The audio is filtered through Silero Voice Activity Detection (VAD) and converted to translated text and spoken audio for remote listeners.
🔊 System Audio Channel (Remote)
Captures what remote participants are saying by intercepting your Windows WASAPI audio output loopback (Stereo Mix / Loopback). This allows Voxxwire to translate their incoming speech into your preferred language in real time.
Automatic Headset & Device Detection
Voxxwire automatically identifies active Bluetooth headsets, USB microphones, and default Windows playback devices. If you use external audio interfaces or virtual cables (like VB-Audio Cable), you can manually select input channels from the Settings panel.
3. Whisper ASR Model Benchmarks & Selection
Voxxwire uses OpenAI's Whisper model via faster-whisper (CTranslate2 execution engine) for high-accuracy local speech recognition. Choose the model size based on your hardware capabilities:
| Model Size | VRAM / RAM | Relative Speed | Transcription Accuracy | Recommended For |
|---|---|---|---|---|
| Tiny | ~1.0 GB | ⚡⚡⚡ (3x Real-Time) | Good (Clear Audio) | Older CPUs / Basic Laptops |
| Base | ~1.5 GB | ⚡⚡ (2x Real-Time) | Great | Standard Desktop / Quad-Core CPU |
| Small (Default) | ~2.5 GB | ⚡ (Real-Time) | Excellent | Recommended for standard meetings |
| Medium | ~5.0 GB | 0.7x Real-Time (CPU) | Near-Human | Requires NVIDIA GPU (CUDA) |
| Large-v3 | ~8.0 GB | Requires GPU | State-of-the-Art | NVIDIA RTX GPUs (6GB+ VRAM) |
4. Offline Translation & Language Model Setup
Voxxwire translates text using Argos Translate, an open-source neural machine translation engine that runs offline on OpenNMT-py models. Over 49+ language pairs are supported, including English, Spanish, French, German, Japanese, Hindi, Mandarin Chinese, Russian, Italian, Portuguese, and Korean.
When you select a new language pair for the first time, Voxxwire downloads the lightweight offline translation package (~30MB) once. After initial setup, all translation calculations execute 100% locally on your CPU.
5. Acoustic Feedback Prevention & Timing Gate
A common challenge in real-time desktop translators is acoustic feedback loops — where translated spoken output (TTS) played through system speakers gets re-captured by the loopback channel and translated back to you!
Voxxwire solves this with an intelligent Timing Gate. When Voxxwire plays translated TTS audio, the system loopback capture stream is temporarily suppressed for the exact duration of the playback plus a 200ms buffer, preventing self-reinforcing audio loops.
6. Troubleshooting Common Issues
No audio is being transcribed or translated?
Verify that the correct Microphone and System Audio devices are selected in the left settings panel. Ensure your microphone is unmuted in Windows Settings and that Voxxwire has permission to access your audio input.
Audio transcription feels laggy or delayed?
If running on CPU, switch the Whisper Model size from Small to Base or Tiny in the ASR Model settings section. This drastically reduces CPU load and latency.
How to use Voxxwire with Headphones?
Connect your headphones before launching Voxxwire. Voxxwire auto-detects headset devices and routes system audio loopback seamlessly.