https://github.com/stefan-schmidbauer/voicesnip
VoiceSnip - Push-to-Talk Speech-to-Text for Linux
https://github.com/stefan-schmidbauer/voicesnip
faster-whisper linux push-to-talk speech-to-text whisper windows
Last synced: 2 months ago
JSON representation
VoiceSnip - Push-to-Talk Speech-to-Text for Linux
- Host: GitHub
- URL: https://github.com/stefan-schmidbauer/voicesnip
- Owner: Stefan-Schmidbauer
- License: mit
- Created: 2025-12-05T15:56:44.000Z (7 months ago)
- Default Branch: main
- Last Pushed: 2026-05-02T09:31:19.000Z (2 months ago)
- Last Synced: 2026-05-02T11:23:40.518Z (2 months ago)
- Topics: faster-whisper, linux, push-to-talk, speech-to-text, whisper, windows
- Language: Python
- Homepage:
- Size: 558 KB
- Stars: 2
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# VoiceSnip - Push-to-Talk Speech-to-Text
Local push-to-talk speech-to-text for Linux. Hold a hotkey, speak, release - text appears at your cursor.
All processing happens locally on your GPU. No cloud, no server, no data leaves your machine.
## Features
- **Push-to-Talk**: Hold hotkey to record, release to transcribe
- **Local GPU Processing**: Whisper runs directly on your NVIDIA (CUDA) or AMD (ROCm) GPU
- **Privacy-First**: All data stays on your device
- **Configurable Hotkeys**: Any key combination (Ctrl+Space, Alt+R, etc.)
- **Multi-Language**: 10 languages (German, English, French, Spanish, etc.) + Auto-Detection
- **Dark/Light Mode**: Switch between dark and light themes
- **Adjustable Font Size**: A-/A+ buttons to customize text size
## Requirements
- Linux (X11 or Wayland)
- Python 3.8+
- NVIDIA GPU with CUDA **or** AMD GPU with ROCm
- Debian/Ubuntu for automated system package install
## Installation
```bash
git clone https://github.com/Stefan-Schmidbauer/voicesnip.git
cd voicesnip
./install.py
```
The installer will ask you to choose a profile:
| Profile | Description |
| -------- | -------------------------------------------------------- |
| **cuda** | Whisper with NVIDIA GPU acceleration (requires CUDA) |
| **rocm** | Whisper with AMD GPU acceleration (requires ROCm, Linux) |
### GPU Setup
**NVIDIA (CUDA):** Install [NVIDIA drivers](https://developer.nvidia.com/cuda-downloads). CUDA libraries are installed automatically by the installer.
**AMD (ROCm):** Install [ROCm drivers](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/) first. PyTorch ROCm is installed automatically by the installer.
## Usage
```bash
./start.sh
```
1. Select your microphone
2. Click "Start"
3. Hold your hotkey (default: Ctrl+Space) and speak
4. Release - text appears at your cursor
## Wayland Support
VoiceSnip works on both X11 and Wayland, but with different workflows due to Wayland's security model.
### X11 (Full Support)
- **Global hotkeys work** - Hold Ctrl+Space (or your configured hotkey) anywhere
- Text is automatically inserted at your cursor via xdotool
### Wayland (GUI-Based Recording)
Wayland blocks global keyboard hooks for security reasons. VoiceSnip provides an alternative:
1. Click **"Start Recording"** in the GUI (instead of hotkey)
2. Speak, then click **"Stop Recording"**
3. Enable **"Auto-copy to clipboard"** for automatic copying
4. Switch to your target app and **Ctrl+V** to paste
## Whisper Models
| Model | Size | VRAM | Notes |
| -------- | ------ | ------ | ----------------------- |
| tiny | 78 MB | 230 MB | Fastest, lowest quality |
| base | 149 MB | 330 MB | |
| small | 488 MB | 745 MB | Good balance |
| medium | 1.5 GB | 2 GB | |
| turbo | 1.6 GB | 2.2 GB | **Recommended for GPU** |
| large-v3 | 3.1 GB | 3.9 GB | Best quality |
Models download automatically on first use to `~/.cache/huggingface/`.
## Configuration
Settings are saved automatically in the project directory:
| File | Purpose |
| ----------------------- | -------------------- |
| `.env` | Environment config |
| `voicesnip_config.json` | GUI settings |
| `voicesnip_profile.ini` | Installation profile |
## License
MIT License - see LICENSE file.
**Third-Party:** OpenAI Whisper (MIT), NVIDIA CUDA, AMD ROCm
## Author
Stefan Schmidbauer
---
_Developed with AI assistance (Claude/Anthropic)_