Verified project record

wzpan/wukong-robot

A modular Chinese voice assistant framework for makers and hobbyists to build personalized smart speakers and conversational robots. It provides an end-to-end voice pipeline with offline wake-up, multiple speech recognition and synthesis engines, and ChatGPT integration for personal learning and research.

Project overview

Project type: AI Agent · Workflow · Audio & Speech
Use cases: Documents & Office
Deployment: Refer to project documentation
License: MIT

Best for

Chinese makers and developers seeking a modular, self-hosted framework to experiment with voice interaction pipelines and custom smart speaker hardware.

Key capabilities

Provides a full voice interaction pipeline including ASR recognition, NLU parsing, skill matching, and TTS speech synthesis.
Supports offline voice command wake-up via Porcupine and snowboy engines, as well as Muse brain-computer interface and Unihboard shake wake-up.
Features highly modular functional plugins, voice recognition, and speech synthesis components maintained separately for easy inheritance and development.
Provides a companion backend to enable remote control, configuration modification, and log viewing.

Limitations and risks

Only supports Python >= 3.7 and < 3.10, does not support Python 2.x.
Does not support Mac M1 chip.
Default API configurations have usage frequency and concurrency limits requiring personal registration.
Designed strictly for personal learning and research; author takes no responsibility for losses caused by usage.

Getting started

Setup requires obtaining API keys and editing configuration files for speech and conversation services. Users can initiate the system by running python3 wukong.py, waking it with 'snowboy', and accessing the backend at http://localhost:5001.

Evidence and sources

README: wukong-robot 是一个简单、灵活、优雅的中文语音对话机器人/智能音箱项目，目的是让中国的 Maker 和 Haker 们也能快速打造个性化的智能音箱。
README: 目的是让中国的 Maker 和 Haker 们也能快速打造个性化的智能音箱。
README: wukong-robot 是一个简单、灵活、优雅的中文语音对话机器人/智能音箱项目
README: 截至 2023 年 3 月 31 日，wukong-robot 的安装设备数已超过 13,000 台，唤醒次数累积超过了 700,000 次。
README: wukong-robot 被唤醒后，用户的语音指令先经过 ASR 引擎进行 ASR 识别成文本，然后对识别到的文本进行 NLU 解析，再将解析结果进行技能匹配，交给适合处理该指令的技能插件去处理。插件处理完成后，得到的结果再交给 TTS 引擎合成成语音，播放给用户。

Reviewed at: 2026-07-16T23:42:17Z