Skip to content

Latest commit

 

History

History
170 lines (137 loc) · 5.54 KB

FEATURES.md

File metadata and controls

170 lines (137 loc) · 5.54 KB

Whisper Dictation Features

Core Functionality

Speech-to-Text

  • 100% Local Processing - No cloud, no telemetry, works offline
  • Push-to-Talk Interface - Hold Super+Period, speak, release to paste
  • Real-time Feedback - GTK4 desktop notifications show status
  • Universal Compatibility - Works in any Wayland application

Language Support

  • 99+ Languages - Full Whisper language support
  • Auto-Detection - Automatically detects language (Italian, English, Spanish, French, etc.)
  • Manual Selection - Force specific language for fastest processing
  • Mixed Language - Auto-detect handles seamless language switching

Performance Options

Model Size Speed Accuracy Use Case
tiny 39 MB ~1-2s ⚡ 60% Quick notes, testing
base 142 MB ~4-6s ⚡⚡ 70% Recommended for speed
small 466 MB ~10-15s 80% Balanced performance
medium 1.5 GB ~20-30s 85% High accuracy
large 2.9 GB ~40-60s 90% Maximum accuracy

Developer Features

Quality Gates

  • Code Formatting - Black + Ruff auto-formatting
  • Complexity Checks - CCN < 10 enforced
  • Security Scanning - Gitleaks + Semgrep
  • Test Coverage - >75% requirement
  • Pre-commit Hooks - Automatic quality enforcement

Development Environment

  • DevEnv Integration - Pure, reproducible development shell
  • Cursor AI Rules - Pre-configured code intelligence
  • Convenience Commands - run-daemon-auto, run-daemon-en, etc.
  • Debug Mode - Comprehensive logging with run-daemon-debug

Privacy & Security

Privacy-First Design

  • No network requests during transcription
  • Audio files deleted after processing
  • No keyboard event logging (only hotkey detection)
  • Local model storage (no external dependencies)

Security Features

  • Virtual device filtering (excludes ydotoold, xdotool)
  • Input group permission requirements
  • Subprocess hardening
  • Secret scanning in git commits

Customization

Hotkey Configuration

  • Customize key combination (default: Super+Period)
  • Multiple modifier key support (Super, Ctrl, Alt, Shift)
  • Per-device keyboard selection

Post-Processing

  • Remove filler words ("um", "uh", etc.)
  • Auto-capitalization
  • Auto-punctuation (experimental)

UI Themes

  • Dark mode (default)
  • Light mode
  • Auto (follows system theme)

Integration

System Integration

  • Wayland-native (ydotool)
  • GNOME desktop notifications
  • PulseAudio/PipeWire audio capture
  • evdev keyboard monitoring

Development Integration

  • NixOS flakes support
  • DevEnv/direnv compatibility
  • Git hooks with Commitizen
  • Cursor AI workspace rules

Command Reference

Daily Usage

run-daemon-auto      # Auto-detect language
run-daemon-en        # English only
run-daemon-it        # Italian only
run-daemon-debug     # Debug mode

Development

test                 # Run test suite
format               # Format code
quality-check        # Run all gates
setup-dev            # First-time setup

Configuration

dictate-en           # Switch to English
dictate-it           # Switch to Italian
vim ~/.config/whisper-dictation/config.yaml

Roadmap

Planned Features

  • Streaming transcription (live preview)
  • Custom vocabulary training
  • Voice command mode
  • LLM API integration (Claude, GPT-4)
  • Multi-backend support (Avalon, Deepgram)
  • Voice profiles for different contexts
  • GPU acceleration
  • systemd service auto-start

Potential Enhancements

  • Punctuation learning
  • Speaker diarization
  • Background noise filtering
  • Custom hotkey actions
  • Text-to-speech feedback
  • Multi-device sync

Comparison to Alternatives

Feature Whisper Dictation Aqua Voice Talon Voice
Privacy ✅ 100% Local ❌ Cloud ✅ Local
Cost ✅ Free 💲 $8/mo 💲 $15/mo
NixOS Support ✅ Native ❌ No ⚠️ Manual
Speed (base) ⚠️ 4-6s ✅ 850ms ✅ 1-2s
Technical Terms ⚠️ 70% ✅ 97% ✅ 95%
Wayland ✅ Yes ⚠️ Limited ❌ X11 only
Multilingual ✅ 99 languages ✅ Yes ⚠️ Limited
Offline ✅ Yes ❌ No ✅ Yes
Open Source ✅ MIT ❌ No ❌ No

Known Limitations

Current Limitations

  • Speed slower than commercial solutions (4-6s vs 850ms for Aqua Voice)
  • Technical jargon accuracy varies by model (70-90%)
  • CPU-only processing (GPU support planned)
  • Single-user design (no multi-user profiles yet)
  • No streaming transcription (processes after key release)

System Requirements

  • NixOS or Linux with Nix
  • Wayland compositor (X11 may work with modifications)
  • Minimum 4GB RAM (8GB recommended for medium model)
  • User must be in input group
  • PulseAudio or PipeWire

Workarounds

  • Use base model for faster processing (trade-off: lower accuracy)
  • Commercial alternatives (Aqua Voice, Avalon API) for maximum speed
  • GPU acceleration can reduce times by 5-10x (implementation pending)

For detailed setup instructions, see README.md For development guide, see DEVELOPMENT.md