LocalWisp 🎙️

LocalWisp is an offline voice dictation utility for Windows. It allows you to dictate text directly into any active application by holding a shortcut, with all audio processing occurring entirely on your local CPU.

Features

• Privacy First: 100% offline; no audio or text data is ever transmitted over the internet.
• CPU Inference: Optimized for multi-core processors using faster-whisper and int8 quantization.
• High Accuracy: Powered by the large-v3-turbo model and silero-vad for robust speech detection.
• Minimalist HUD: A semi-transparent overlay provides real-time status feedback (Recording/Processing).
• Direct Injection: Automatically pastes the transcribed text at your current cursor position.

Hardware & Current Optimization

The utility is currently tuned for a specific performance profile based on the developer's hardware:

• Reference CPU: AMD Ryzen 7 5700G (8 Cores / 16 Threads).
• Thread Management: Hardcoded to 8 CPU threads to ensure fast processing without system lag.
• Memory: 32 GB RAM recommended for the large-v3-turbo model footprint.

Future Development & Collaboration

LocalWisp is an open project. While built for a specific setup, future iterations aim to:

• Implement automatic hardware detection to dynamically set thread counts.
• Improve flexibility for different model sizes based on available system resources.
• We welcome collaborative efforts to improve hardware abstraction and performance.

Installation & Usage

1. Install dependencies: pip install -r requirements.txt.
2. Setup models: python setup_models.py.
3. Run: Use Launch_LocalWisp.bat to start the tool in the background.
4. Operation: Hold Win + Ctrl to record. Release the keys to transcribe and paste.

License

Copyright (c) 2026 ATTrindade and LocalWisp Contributors. Distributed under the MIT License.