CapsWriter-Offline is a speech-to-text input tool for Windows that operates entirely offline with local speech recognition.
The tool solves the problem of needing fast, accurate voice input without relying on cloud services or internet connectivity. It works by holding down the CapsLock key or a mouse side button to record speech, then releasing to insert the recognized text directly into any application. The core approach combines multiple speech recognition engines with local processing, offering options ranging from fast lightweight models to more accurate larger ones. It includes features for handling numbers, custom vocabulary through hot-word replacement, regex-based text substitution, and LLM-based post-processing roles that can refine or transform recognized text. The tool also supports transcribing audio and video files to subtitles, text, and timestamped JSON output.
Developers should choose this tool if they work on Windows and need reliable offline speech input without cloud dependencies or privacy concerns. It suits workflows where low latency matters and customization is valuable, such as documentation, coding with voice commands, or transcription work. The project distinguishes itself through its complete offline operation, support for multiple recognition engines with different accuracy-speed tradeoffs, and extensive customization options via hot-word files and regex rules. It requires no installation and can run from a USB drive, making it suitable for restricted or air-gapped environments.
The project shows active development with comprehensive documentation covering environment setup, feature configuration, language support, GPU acceleration options, and model selection. Multiple recognition engines are supported with documented performance characteristics, and the tool includes a changelog tracking updates. The codebase is written in Python and includes both server and client components, allowing flexibility in deployment even on older systems that cannot run the recognition models locally.
