Whisper Key - Local Speech-to-Text

Global hotkeys to record speech and transcribe directly to your cursor.

Questions or ideas? Discord

✨ Features

• Global Hotkey: Start recording speech from any app
• Auto-Paste: Transcribe directly to cursor
• Auto-Send: Optionally auto-send with ENTER keypress
• Local/Offline: Voice data never leaves your computer
• CPU Ready: Small, efficient models available
• GPU Ready: Support for both NVIDIA & AMD cards
• Cross-platform: Works on Windows and macOS
• Voice Commands: Trigger shortcuts, text snippets, and shell commands by voice — docs
• Configurable: Customize hotkeys, models, and much more

🚀 Quick Start

From PyPI (Recommended)

Requires Python 3.11-3.13

# With pipx (isolated environment)
pipx install whisper-key-local

# Or with pip
pip install whisper-key-local

Then run: whisper-key (or wk for short)

Windows App

1. Download whisper-key.exe from the latest release
2. Run whisper-key.exe

From Source

git clone https://github.com/PinW/whisper-key-local.git
cd whisper-key-local
pip install -e .
python whisper-key.py

🎤 Basic Usage

Hotkey	Windows	macOS
Start recording	Ctrl+Win	Fn+Ctrl
Stop & transcribe	Ctrl	Fn
Stop & auto-send	Alt	Option
Cancel recording	Esc	Shift
Voice command mode	Alt+Win	Fn+Command

Open the system tray / menu bar icon to:

• Toggle auto-paste vs clipboard-only
• Change transcription model
• Select audio device

🗣️ Voice Commands

Speak trigger phrases to run shell commands and more. Define in:

• Windows: %APPDATA%\whisperkey\commands.yaml
• macOS: ~/.whisperkey/commands.yaml

commands:
  # Send a keyboard shortcut
  - trigger: "undo"
    hotkey: "ctrl+z"
  # Deliver pre-written text
  - trigger: "my email"
    type: "[email protected]"
  # Run a shell command
  - trigger: "open notepad"
    run: 'notepad.exe'

See the Voice Commands Guide for full details.

⚡ GPU Acceleration

Whisper Key detects your GPU on first launch and offers one-press install of the required runtime libraries. Supports NVIDIA (CUDA) and AMD (ROCm).

For manual setup or troubleshooting, see the GPU Setup Guide.

⚙️ Configuration

Local settings at:

• Windows: %APPDATA%\whisperkey\user_settings.yaml
• macOS: ~/.whisperkey/user_settings.yaml

Delete this file and restart app to reset to defaults.

Option	Default	Notes
Whisper
whisper.model	tiny	Any model defined in whisper.models
whisper.device	cpu	cpu or cuda (NVIDIA/AMD GPU) — setup guide
whisper.compute_type	int8	int8/float16/float32
whisper.language	auto	auto or language code (en, es, fr, etc.)
whisper.beam_size	5	Higher = more accurate but slower (1-10)
whisper.initial_prompt	""	Guide transcription style, language variant, or script
whisper.hotwords	[]	Words the model should favor (names, technical terms)
whisper.models	(see config)	Add custom HuggingFace or local models
Post-Processing
post_processing.strip_trailing_period	false	Strip trailing period from output
post_processing.corrections	{}	Fix recurring misheard words, e.g. CAPEX: [cap x]
Hotkeys
hotkey.recording_hotkey	ctrl+win / fn+ctrl	Windows / macOS
hotkey.stop_key	ctrl / fn	Stop recording
hotkey.auto_send_key	alt / option	Stop + paste + Enter
hotkey.cancel_combination	esc / shift	Cancel recording
hotkey.recording_mode	toggle	toggle or push_to_talk
hotkey.command_hotkey	alt+win / fn+command	Voice command mode
Voice Activity Detection
vad.vad_precheck_enabled	true	Prevent hallucinations on silence
vad.vad_onset_threshold	0.7	Speech detection start (0.0-1.0)
vad.vad_offset_threshold	0.55	Speech detection end (0.0-1.0)
vad.vad_min_speech_duration	0.1	Min speech segment (seconds)
vad.vad_realtime_enabled	true	Auto-stop on silence
vad.vad_silence_timeout_seconds	30.0	Seconds before auto-stop
Audio
audio.host	null	Audio API (WASAPI, Core Audio, etc.)
audio.channels	1	1 = mono, 2 = stereo
audio.dtype	float32	float32/int16/int24/int32
audio.max_duration	900	Max recording seconds (0 = unlimited)
audio.input_device	default	Device ID or "default"
Clipboard
clipboard.auto_paste	true	false = clipboard only
clipboard.delivery_method	paste	paste (Ctrl+V) or type (direct injection)
clipboard.paste_hotkey	ctrl+v / cmd+v	Paste key simulation
clipboard.paste_pre_paste_delay	0.05	Delay after copy, before paste hotkey (seconds)
clipboard.paste_preserve_clipboard	true	Restore clipboard after paste
clipboard.paste_clipboard_restore_delay	0.5	Delay before clipboard restore (seconds)
clipboard.type_also_copy_to_clipboard	false	Also copy to clipboard in type mode
clipboard.type_auto_enter_delay	0.15	Delay before ENTER after typing (seconds)
clipboard.type_auto_enter_delay_per_100_chars	0.1	Extra ENTER delay per 100 typed chars (seconds)
Logging
logging.level	INFO	DEBUG/INFO/WARNING/ERROR/CRITICAL
logging.file.enabled	true	Write to app.log
logging.log_transcriptions	false	Include transcribed text in log (privacy)
logging.console.enabled	true	Print to console
logging.console.level	WARNING	Console verbosity
Audio Feedback
audio_feedback.enabled	true	Play sounds on record/stop
audio_feedback.transcription_complete_enabled	false	Play sound on transcription complete
audio_feedback.ready_enabled	true	Play sound when app finishes loading
audio_feedback.start_sound	assets/sounds/...	Custom sound file path
audio_feedback.stop_sound	assets/sounds/...	Custom sound file path
audio_feedback.cancel_sound	assets/sounds/...	Custom sound file path
audio_feedback.transcription_complete_sound	assets/sounds/...	Custom sound file path
audio_feedback.ready_sound	assets/sounds/...	Custom sound file path
System Tray
system_tray.enabled	true	Show tray icon
system_tray.tooltip	Whisper Key	Hover text
Terminal Title
terminal_title.idle	""	Tab title prefix when idle: static string or [prefix, seconds] animation frames
terminal_title.recording	🔴 blink	Tab title prefix while recording
terminal_title.processing	""	Tab title prefix while transcribing
Console
console.start_hidden	false	Hide console after startup (whisper-key-hideable.exe only)
Update
update.mode	prompt	prompt or auto
Voice Commands
voice_commands.enabled	true	Enable voice command mode

📁 Model Cache

Default path for transcription models (via HuggingFace):

• Windows: %USERPROFILE%\.cache\huggingface\hub\
• macOS: ~/.cache/huggingface/hub/

Contributing

Check the roadmap for planned features and see CONTRIBUTING.md for guidelines. Please open an issue before starting work on new features.

📦 Dependencies

Cross-platform: faster-whisper · numpy · sounddevice · soxr · pyperclip · ruamel.yaml · pystray · Pillow · playsound3 · ten-vad · hf-xet

Windows: global-hotkeys · pywin32

macOS: pyobjc-framework-Quartz · pyobjc-framework-ApplicationServices