Repolode
Submit
HaujetZhao

HaujetZhao/CapsWriter-Offline

Offline Windows voice input tool: hold CapsLock or mouse side button X2 to speak, release to type the recognized text.

Open-source alternative to:Dragon NaturallySpeaking

View on GitHub
Stars
6,985
Forks
653
Open issues
216
Last commit
last month

What is it?

CapsWriter-Offline is a fully offline voice input tool for Windows. Hold CapsLock or mouse side button X2 to speak, and the recognized text is typed into the active window when you release the key. It supports push-to-talk and single-click recording modes, and runs as a client/server pair so that the speech models can run on a separate machine.

The project offers several recognition engines, including Paraformer, SenseVoice-Small, Fun-ASR-Nano and Qwen3-ASR, some of which can use GPU acceleration. Beyond live dictation, it can transcribe audio and video files into .srt subtitles, .txt text and .json timestamps. Text post-processing includes Chinese number conversion, hot-word replacement by phonetic fuzzy matching, regex replacement rules, and LLM-based roles that process output when it begins with a role name.

The software targets Windows 10/11 (64-bit). Linux and macOS are not supported, and the client can run on older Windows 7 machines while the server model cannot.

Key features

  • 01Push-to-talk input with CapsLock or mouse side button X2, typing on release.
  • 02Offline speech recognition with multiple selectable engines, some with GPU acceleration.
  • 03Transcribes audio and video files into SRT subtitles, text and timestamped JSON.
  • 04Hot-word replacement via phonetic fuzzy matching, plus regex replacement rules.
  • 05Converts spoken Chinese numbers into Arabic numerals.
  • 06Client/server split so the client can run on older machines.

Use cases

  • Dictating text into any Windows application without sending audio to the cloud.
  • Working on machines that must stay offline, such as secured or air-gapped computers.
  • Generating subtitles and transcripts from local audio or video files.
  • Maintaining a custom vocabulary of rare terms so they are recognized correctly.

Quick start

  1. 1

    Install VC++ runtime (and ffmpeg for file transcription)

  2. 2

    Download the latest release and model archive, unzip models into the models folder

  3. 3

    Double-click start_server.exe to start the server

    start_server.exe
  4. 4

    Double-click start_client.exe, then hold CapsLock or mouse side button X2 to speak

    start_client.exe

Things to know

  • Officially supported only on Windows 10/11 (64-bit); Linux is untested and macOS is not supported.
  • Requires separately downloading model files before it works.
  • Server models can't run on Win7 machines; those can only use the client.

Topics

#speech-recognition#voice-input#offline#windows#asr#subtitles#whisper-alternative#hot-words

FAQ

What is CapsWriter-Offline?
Offline Windows voice input tool: hold CapsLock or mouse side button X2 to speak, release to type the recognized text. CapsWriter-Offline is a fully offline voice input tool for Windows. Hold CapsLock or mouse side button X2 to speak, and the recognized text is typed into the active window when you release the key. It supports push-to-talk and single-click recording modes, and runs as a client/server pair so that the speech models can run on a separate machine.
Is CapsWriter-Offline free to use? What license does it have?
CapsWriter-Offline is open source under the MIT license. Check the license terms before using it in commercial products.
Is CapsWriter-Offline still maintained?
Updated within 6 months. The last commit to CapsWriter-Offline was on Sep 14, 2026.
What can I use CapsWriter-Offline for?
Dictating text into any Windows application without sending audio to the cloud. Working on machines that must stay offline, such as secured or air-gapped computers. Generating subtitles and transcripts from local audio or video files. Maintaining a custom vocabulary of rare terms so they are recognized correctly.
How do I install CapsWriter-Offline?
Install VC++ runtime (and ffmpeg for file transcription) Download the latest release and model archive, unzip models into the models folder Double-click start_server.exe to start the server: start_server.exe Double-click start_client.exe, then hold CapsLock or mouse side button X2 to speak: start_client.exe
Where can I get CapsWriter-Offline?
The source code and installation instructions are on GitHub: https://github.com/HaujetZhao/CapsWriter-Offline

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects