Digital human based on ESP32S31

Created an interactive little crocodile digital human using ESP32-S31, fully embedded and locally deployed, offering an ultra‑healing experience ✨

After connecting to Wi‑Fi, it interfaces with AI via WebSocket to enable voice conversation + TTS playback, with the screen simultaneously showing chat content and the digital human’s status. In addition, a companion Android App was created for the ESP32‑S31 device, mainly addressing two issues: making network provisioning for the device more convenient, and letting the device know “I’m nearby.”

The significance of this approach is:

The device is no longer just a piece of hardware that “waits for user operation,” but can integrate more naturally into usage scenarios. After the user initially uses the App to configure the network and bind the device, thereafter simply bringing the phone close allows the device to greet proactively, without needing to press a button each time, reconnect Bluetooth, or manually wake it.

✅ Core Features

- WiFi provisioning + one‑click reconnection after disconnection

- Real‑time microphone capture, AI text echo, TTS playback

- Supports Chinese + emoji screen display

- Four‑state switching: standby / listening / thinking / broadcasting

- Volume‑driven dynamic mouth shapes, synchronized with speech rhythm

💡 Core Optimization Highlights

Abandoning a crude geometric face, we use a layered image design, separating the crocodile base, mouth shapes, expressions, and blink layers, converting them into LVGL‑compatible C resources. This eliminates the need for the ESP32 to decode PNGs in real time, greatly saving device performance and resulting in smoother animations.

Optimized five mouth positions (closed / slightly open / medium open / wide open / O‑shape), dynamically switching in sync with TTS audio, giving a realistic speaking impression!

📝 Project Summary

A thoroughly complete embedded AI hands‑on project, covering audio capture, network communication, UI display, digital human animation, and more. Thanks to the Espressif development board for enabling the implementation!