Offline wake word
The VB6824 offline voice engine detects the wake word on-device; audio is only streamed when the device is addressed.
Smart Home & AI / Solution
Offline voice wake, dual-band Wi-Fi 6 / 4G and a round display for AI toys, companion robots and smart-home devices.
Start a brief01Overview
A reference module for AI toys and companion devices: wake word on-device, then a cloud or local LLM for conversation, expressions and device actions.
The ULINK-M33 kit contains six parts: ESP32-C5 core board, 2.0-inch round display, 4G Cat.1 module, microphone, 1200 mAh battery and 8 Ω 1 W speaker. It supports offline voice wake (VB6824), dual-band Wi-Fi 6, BLE 5.4 and 4G Cat.1.
The module can access 小智 AI and cloud LLM providers such as Doubao, DeepSeek, ChatGPT and Qianwen, with changeable roles and an expression UI. GPIO, audio and Type-C interfaces support sensors, extra displays and OTA updates; the same architecture can be adapted to smart-home devices.

02Key features
Each feature should map to a specification, a test or a documented application — not a generic marketing claim.
The VB6824 offline voice engine detects the wake word on-device; audio is only streamed when the device is addressed.
ESP32-C5 provides dual-band Wi-Fi 802.11b/g/n/ax and BLE 5.4; the NT26 module adds LTE Cat.1 for mobile use.
A 240 × 296 round display for eyes, expressions, status and a simple UI without a phone.
Works with 小智 AI and cloud providers including Doubao, DeepSeek, ChatGPT and Qianwen, with changeable character roles.
Core board, display, 4G module, microphone, 1200 mAh battery and 8 Ω 1 W speaker can be assembled and serviced separately.
GPIO, audio and Type-C interfaces support sensors, extra displays, programming and firmware updates.
03Working principle
The microphone captures speech, the VB6824 engine detects the wake word, and the ESP32-C5 streams audio through Wi-Fi or 4G to the selected LLM; the reply drives speech and the round display.
The offline engine detects the wake word and starts the interaction without sending continuous audio.
ESP32-C5 uses dual-band Wi-Fi 6 or the 4G Cat.1 module to reach the cloud service.
ASR, LLM and TTS run in the cloud (or on a local host); the module receives text/audio and an emotion state.
The 8 Ω 1 W speaker plays the reply and the round display shows expressions; GPIO can drive motion or lights.
04Model selection
Compare the series against the requirement. Every value should state its unit and test condition.
Reference ULINK-M33 configuration from the supplied product sheet. Other controllers, display sizes and radios are available per project.
| Model | Controller | Wireless | Display | Audio | Battery |
|---|---|---|---|---|---|
| ULINK-M33 | ESP32-C5 | Wi-Fi 6 2.4/5 GHz + BLE 5.4 + 4G Cat.1 | 2.0 in / 240 × 296 | Mic -32 ± 3 dB, 8 Ω 1 W | 1200 mAh |
| Custom | ESP32 / Linux-class | Wi-Fi / BLE / 4G | 2.0–2.8 in | Custom | 1200–3000 mAh |
05Applications
Add the operating requirement that makes this product a good match — not just an industry logo.
Round display, voice wake and LLM conversation for plush toys and companion robots.
Add wake-word and conversational control to a speaker, display or home hub.
Voice control for kitchen, laundry, lighting and HVAC products without a full screen.
Room control and assistance requests with privacy modes and local fallback.
06Custom development
The ULINK-M33 is a reference kit. The same architecture can be developed around your product: a custom board, firmware, mobile app and cloud service, with certification support.
07Compliance & quality
Certificate scope and test reports are supplied on request. “Project-specific” means the standard is confirmed for the selected configuration; “pending” means it has not yet been verified for publication.
Design and documentation support; final scope depends on the selected radio module and enclosure.
Wake-only capture, hardware mute and local-only options are part of the module specification.
08Downloads
Datasheets and drawings can be shared directly with your engineering and procurement teams.
Next step
Send the application, target volume and market for Voice Wake Module + LLM Interface. We will confirm stock availability or scope a custom build, with samples and a production plan.