Feature Usage Tutorial
This page summarizes what XiaoZhi can do, following the “Features Implemented” list of the upstream README for firmware v2.5.2. Exact capabilities depend on the board hardware (display, camera, echo cancellation, and so on), so check the notes in your board’s directory.
Voice Dialogue
- Offline wake word: based on ESP-SR, with custom wake words supported.
- Two dialogue approaches: the traditional streaming ASR + LLM + TTS pipeline, or a Realtime end-to-end speech model. Audio is streamed as Opus.
- Full duplex: hardware with AEC (acoustic echo cancellation) can hold real-time full-duplex conversations.
- Speaker recognition: based on 3D Speaker, it identifies the current speaker.
- Protocols: WebSocket or MQTT + UDP, chosen by the server in its OTA response; see First Boot.
Display and Feedback
- OLED / LCD displays with expressions and rich emotion rendering; see Emotion display.
- Battery level display and power management.
- Camera vision input on some hardware.
- The upstream README lists 38 interface languages. Voice prompts use localized resources first and fall back to English when they are missing.
Device Control (MCP)
- On-device MCP: control volume, lights, motors, GPIO, and more.
- Cloud MCP: extend the model with smart-home control, PC desktop operation, knowledge search, email, and more.
See the MCP documentation for the protocol and usage.
Personalization
Custom wake words, fonts, emotions, and chat backgrounds can be edited online in a web page using the upstream Assets generator.
Connectivity
Wi-Fi, wired Ethernet, USB RNDIS, and Cat.1 4G are supported, and Wi-Fi can be set up through a hotspot or BluFi. See the Network Configuration Guide.
Source: the “Features Implemented” section of README.md in 78/xiaozhi-esp32 v2.5.2.