A desktop application that allows you to record audio and transcribe it using OpenAI's Whisper model. The application runs locally and can be triggered using keyboard shortcuts.
- Record audio using keyboard shortcuts
- Transcribe audio using Whisper.cpp
- Automatic text pasting after transcription
- Desktop notifications for recording status
- Configurable audio settings and keyboard shortcuts
- Install Rust and Cargo from rustup.rs
- (Optional) Install the required system dependencies:
- ALSA development libraries (for audio recording)
- X11 development libraries (for keyboard input)
- OpenSSL development libraries (for model downloading)
- Build and run:
cargo run --release --features metal- Build and run:
cargo run --release --features cuda,x11 # Untested
cargo run --release --features cuda,wayland# Clone the repository
git clone https://github.com/yourusername/whispering.git
cd whispering
# Build and run using Nix
nix buildThe application can be configured using a TOML file located at ~/.config/whispering/config.toml. If no configuration file exists, the application will create one with default values.
# Whispering Configuration Example
#
# Copy this file to ~/.config/whispering/config.toml and modify the values as needed.
# If no config.toml is found, the application will use default values.
[audio]
# Number of audio channels (1 for mono, 2 for stereo)
channels = 1
# Sample rate in Hz
sample_rate = 16000
# Bits per sample
bits_per_sample = 32
# Sample format (float or int)
sample_format = "float"
[model]
# Hugging Face model repository
repo = "ggerganov/whisper.cpp"
# Model filename
filename = "ggml-base.en.bin"
# Prompt
# Options are `type = "none", "vocabulary", "raw"
# prompt = { type = "vocabulary", vocabulary = ["Google", "HuggingFace"] }
# will insert "Google, HuggingFace" as an initial prompt.
#
# prompt = { type = "raw", prompt = "HuggingFace likes Google"] }
# will insert "HuggingFace like Google" as an initial prompt.
#
# For more information on whisper prompting:
# https://cookbook.openai.com/examples/whisper_prompting_guide
prompt = { type = "none" }
replacements = {}
[paths]
# Cache directory for storing temporary files
cache_dir = "~/.cache/whispering"
# Path to the recorded audio file
recording_path = "~/.cache/whispering/recorded.wav"
[shortcuts]
# Keys that need to be pressed in sequence to start recording
# Available keys: control, alt, shift, super, space, enter, r, and many others
keys = ["ControlLeft", "Space"]
# Automatically hit enter after sending the text (so sends a message in usual contexts).
autosend = falsechannels: Number of audio channels (1 for mono, 2 for stereo)sample_rate: Sample rate in Hz (default: 16000)bits_per_sample: Bits per sample (default: 32)sample_format: Sample format ("float" or "int")
repo: Hugging Face model repositoryfilename: Model filename to download and use
cache_dir: Directory for storing temporary filesrecording_path: Path to save recorded audio files
keys: List of keys to press in sequence to start recording- Available keys: control, alt, shift, super, space, enter, and many others
- Default: ["control", "space"]
- Start the application
- Press the configured shortcut keys (default: Control + Space) to start recording
- Release the last key to stop recording
- The transcribed text will be automatically pasted into the active window
- Ensure your microphone is properly connected and selected as the default input device
- Check that your system's audio permissions are properly configured
- Verify that the ALSA development libraries are installed
- Make sure the X11 development libraries are installed
- Check that your window manager supports the keyboard shortcuts you've configured
- Ensure you have sufficient disk space for the model (approximately 1.5GB)
- Check your internet connection for model downloading
- Verify that the model repository and filename are correct
This project is licensed under the MIT License - see the LICENSE file for details.