MasterType V5

简体中文

Portable offline voice typing for Windows, built for coding, research, and fast text capture.

Download Latest Release | View Third-Party Notices | 中文说明

Turn a Windows machine into a portable offline dictation workstation with a bundled runtime, configurable hotwords, and a practical client/server workflow.

Why MasterType

• Offline by default: no cloud speech pipeline required for the core workflow
• Portable release: unzip, add models, launch, and use on another Windows PC
• Dual ASR path: Qwen-ASR preferred, Fun-ASR-Nano CPU as fallback
• Fast text workflow: dictation, clipboard paste, direct typing, and cleanup rules
• Hackable package: hotwords, regex replacement, shortcuts, and LLM roles stay editable

Quick Start

1. Download the latest packaged release from GitHub Releases.
2. Extract it anywhere on a Windows machine.
3. Put the required speech models into models/.
4. Run start_server.exe.
5. Run start_client.exe.
6. Press Scroll Lock to start and stop recording.

Product Snapshot

Area	What you get
Runtime	Portable Windows package with bundled dependencies
Input flow	Scroll Lock record toggle, clipboard paste mode, and direct typing
Recognition	Qwen-ASR first, Fun-ASR-Nano CPU fallback
Customization	Hotwords, regex replacement, startup scripts, and role templates
Output	Plain text, subtitles, JSON, and post-processing rules

Release Download

This repository is source-visible, but the main end-user entry point is the packaged Windows release.

• Download the latest packaged build from the Releases page
• Model files are not included in the repository or release zip
• Keep THIRD_PARTY_NOTICES.md when redistributing the package

Release Notes

This release package:

• includes the portable runtime and launch scripts
• does not include speech model files
• keeps a generic AI-research-oriented hotword set
• removes obvious personal and donation-related content

Models

Model files are not bundled.

You can choose one of the following approaches:

1. Download compatible packaged model files from the upstream model release page used by this project: https://github.com/HaujetZhao/CapsWriter-Offline/releases/tag/models
2. Check the official upstream model projects and prepare/convert the required files yourself: https://github.com/alibaba-damo-academy/FunASR https://github.com/QwenLM/Qwen3-ASR

After downloading, place the files under the local models/ directory. See models/README.txt.

Features

• Click Scroll Lock once to start recording, click again to finish
• Clipboard paste mode and direct typing mode
• Portable tray startup and silent startup scripts
• AI-research-oriented default hotwords and correction rules
• Optional LLM roles with blank API keys by default
• File transcription support for subtitles, text, and JSON

Portable Deployment

Recommended steps on another Windows PC:

1. Install Microsoft VC++ runtime if missing.
2. Extract the release zip anywhere.
3. Download model files separately into models/.
4. Run start_server.exe.
5. Run start_client.exe.
6. Optionally run install_startup.vbs for startup integration.

Files Worth Editing

• config_client.py: hotkeys, paste mode, clipboard restore logic
• config_server.py: preferred model and fallback thresholds
• hot.txt: strong replacement hotwords
• hot-rule.txt: regex replacements
• hot-server.txt: soft context hotwords for Fun-ASR

Who This Is For

• Users who want an offline Windows dictation tool without a full installer workflow
• Developers who want to tweak hotwords, text cleanup, and startup behavior
• Researchers who want a practical local speech-to-text package they can move between machines

LLM Roles

The LLM/ folder is included as examples only.

• API keys are blank by default
• you can keep them disabled safely
• if enabled, review each role file before use

Third-Party and Licensing

This release is based on multiple upstream open/public projects and bundled Python runtime dependencies.

Special thanks to the upstream work that made this portable package practical:

• CapsWriter-Offline for the overall offline voice typing workflow, code organization ideas, and model packaging conventions
• FunASR for the fallback ASR path and related runtime integration
• Sherpa-ONNX for offline speech runtime components used in the packaged environment
• Qwen3-ASR for the preferred recognition path and related integration work

Review THIRD_PARTY_NOTICES.md before public redistribution.

If you redistribute or fork this repository publicly, keep that notice file with the package and preserve attribution to the upstream projects listed there.

Public Repository Notes

This package is intended to be practical and hackable rather than fully installer-driven.

• The repository contains the portable Windows runtime for convenience
• Large bundled runtime files are tracked with Git LFS where needed
• Speech model files are intentionally excluded
• Public forks should preserve upstream attribution and notice files