SuperWhisper Linux

AI-powered voice-to-text for Linux, inspired by SuperWhisper. Works with Hyprland and other wlroots-based Wayland compositors.

Features

• 🎤 Push-to-talk voice recording via hotkey
• 🤖 Local transcription using whisper.cpp (no internet required)
• ☁️ Cloud transcription support (OpenAI, Groq, Deepgram)
• 📋 Auto-paste transcribed text to focused application
• 🌍 Multilingual support (Spanish, English, and more)
• 🖥️ System tray integration

Installation

NixOS / Home-Manager (Recommended)

Add to your flake.nix inputs:

{
  inputs = {
    super-whisper-linux = {
      url = "github:facundopanizza/super-whisper-linux";
      inputs.nixpkgs.follows = "nixpkgs";
    };
  };
}

Then in your home-manager config (home.nix):

{ inputs, pkgs, ... }:

{
  home.packages = [
    inputs.super-whisper-linux.packages.${pkgs.system}.default
  ];
}

Nix (without flakes)

nix-env -if https://github.com/facundopanizza/super-whisper-linux/archive/main.tar.gz

Build from source

# Enter development environment
nix develop

# Build release binary
cargo build --release

# Install to ~/.local/bin
cp target/release/super-whisper-linux ~/.local/bin/

Quick Start

1. Download a Whisper model

# Using the built-in download command (recommended)
super-whisper-linux download-model --model base

# Or manually with curl
mkdir -p ~/.local/share/super-whisper-linux/models
curl -L https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin \
  -o ~/.local/share/super-whisper-linux/models/ggml-base.bin

Available models (larger = more accurate but slower):

Model	Size	Speed	Accuracy
tiny	75 MB	Fastest	Basic
base	142 MB	Fast	Good ✓
small	466 MB	Medium	Better
medium	1.5 GB	Slow	High
large	3 GB	Slowest	Best

2. Create configuration

super-whisper-linux init-config

3. Set up Hyprland keybind

Add to ~/.config/hypr/hyprland.conf:

# SuperWhisper - Push to talk
bind = SUPER, B, exec, super-whisper-linux trigger toggle

4. Start the app

# Run manually
super-whisper-linux

# Or add to Hyprland autostart
# In hyprland.conf:
exec-once = super-whisper-linux

Usage

1. Press SUPER+B to start recording
2. Speak your text
3. Press SUPER+B again to stop and transcribe
4. Text is automatically pasted to the focused application

Commands

# Start the main application
super-whisper-linux

# Trigger commands (when app is running)
super-whisper-linux trigger toggle  # Toggle recording
super-whisper-linux trigger start   # Start recording
super-whisper-linux trigger stop    # Stop and transcribe
super-whisper-linux trigger cancel  # Cancel recording

# Model management
super-whisper-linux download-model --model base  # Download a model

# Utilities
super-whisper-linux devices         # List audio devices
super-whisper-linux status          # Check app status
super-whisper-linux init-config     # Generate config file

Configuration

Edit ~/.config/super-whisper-linux/config.toml:

[general]
default_provider = "whisper-local"  # or "openai", "groq", "deepgram"
language = "auto"                   # or "en", "es", etc.
auto_paste = true                   # Paste after transcription

[audio]
sample_rate = 16000
max_duration = 300                  # Max recording seconds

[providers.whisper-local]
model = "base"                      # tiny, base, small, medium, large

# For cloud providers, set API keys:
# [providers.openai]
# api_key = "sk-..."

Troubleshooting

App not responding to hotkey

• Check if the socket exists: ls $XDG_RUNTIME_DIR/super-whisper.sock
• Check if app is running: super-whisper-linux status

Paste not working

• Ensure wtype is installed and in PATH
• Try pasting manually after transcription (text is copied to clipboard)

No audio recorded

• Check microphone: super-whisper-linux devices
• Test with: arecord -d 3 test.wav && aplay test.wav

Development

# Enter dev environment
nix develop

# Run with debug logging
cargo run -- --debug

# Watch for changes
cargo watch -x run

License

MIT