AI Voice Chips & Modules

The WT2606A can be integrated through UART serial communication. It provides a standard UART control protocol and supports control commands for wake-up, command words, local playback, and more. The peripheral circuit is extremely simple—only MIC, speaker, UART TX/RX, and power connections are required, with no complex secondary development. The customer’s existing MCU can receive voice commands through UART to enable voice control.

The M06 module supports three network provisioning methods: ① BLE Bluetooth quick provisioning (recommended, completed with one tap in the user’s mobile APP) ② WiFi SmartConfig provisioning ③ AP hotspot provisioning. The provisioning success rate is over 99%, with an initial provisioning time of less than 15 seconds. BLE provisioning is also compatible with existing APPs and OTA updates, providing strong IoT scalability.

The M08 module uses a standard Nano-SIM card and supports all three major Chinese mobile networks: China Mobile, China Unicom, and China Telecom. It features a built-in ceramic antenna for stable connectivity even in complex network environments. Data costs are calculated based on device activation volume. Typical voice interaction data usage is 30–50 MB per month, with annual data costs of less than RMB 10 per device. The module also supports packet-loss mitigation algorithms to ensure smooth voice interaction.

An offline recognition model is deployed on the device, supporting basic functions such as wake-up, recognition, and local voice playback when the network is unavailable. Once connected to the Internet, the system automatically switches to cloud-based large language models to enable advanced functions such as multi-turn conversations, semantic understanding, and AI Q&A. The switching between edge and cloud is seamless—dual VAD engines on the device and in the cloud work together without users noticing the transition. For health devices such as blood pressure monitors, voice playback of measurement results remains available even when the network is disconnected.

Yes. Private deployment is supported. The AIoT smart voice platform can be deployed on the customer’s own servers, giving customers full control over device data, voice data, and user data. The platform provides a wide range of communication protocols, including device-side UART, WebSocket, MQTT, and HTTPS, enabling integration with various devices and cloud services. It has passed ISO 27001 information security management system certification.

The platform has pre-integrated mainstream AI large language models and platforms, including DeepSeek, Doubao (Volcano Engine), iFLYTEK AIUI, Coze, and JoyInside (JD.com). Customers can also integrate their specified large language models through open APIs. The platform supports multi-model switching, allowing the same device to use different models according to application scenarios, such as DeepSeek for health consultation and iFLYTEK for educational content.

Wake words can be fully customized, with support for up to 6 wake words. Up to 200 offline command words can also be customized. The customization process is as follows: the customer provides the wake words and command word text → the platform trains the model → the customized model is delivered to the device via OTA. The process usually takes 2–3 business days. Wake words in regional dialects can also be customized, such as Cantonese and Sichuan dialect.

The platform covers 35+ official languages from countries and regions, including Chinese, English, Japanese, Korean, French, German, Spanish, Russian, Portuguese, Italian, Thai, Vietnamese, Indonesian, Malay, Arabic, Turkish, Polish, Czech, Swedish, Dutch, Greek, Ukrainian, Kazakh, Persian, Hindi, Bengali, Urdu, Swahili, Hausa, Tamil, Mongolian, Bulgarian, Romanian, Uzbek, and Filipino. Overseas products require deployment on overseas servers. For specific international applications, please contact technical support. The platform also supports 200+ Chinese regional dialects.

The online TTS engine supports 16 languages, including Chinese, English, Chinese-English mixed speech, French, Thai, Italian, Korean, Japanese, Indonesian, German, Spanish, Russian, Vietnamese, Malay, and Filipino, as well as 7 Chinese regional dialects, including Cantonese, Northeastern Mandarin, Tianjin dialect, Taiwanese Mandarin, Sichuan dialect, and Hunan dialect. It includes highly realistic voices such as Xiao Wei, Beijing-style Kanyeye, Sisi, and Wei Xiaobao. Voice cloning is supported, allowing family members’ voices to be cloned to make voice announcements from blood pressure monitors, children’s toys, and other products feel more personal and friendly. Streaming synthesis latency can be as low as 300 ms.

WT2606A/M06 natively supports a resolution of 480×480, 30 fps animation, built-in 2D graphics acceleration, and QSPI/SPI display interfaces. WT3000A-M08 supports 360×360. LVGL UI framework is supported for rapid development of visual user interfaces. The solution can be adapted to various small color displays without an external display driver IC, providing smooth graphics, lower power consumption, and a simplified peripheral circuit.

The M06 module supports remote OTA updates via WiFi, M08 supports remote OTA updates via 4G, and the WT2606A chip supports UART firmware updates. Silent updates are supported—the firmware can be automatically downloaded in the background and installed automatically when the device is idle, without interrupting the user. OTA supports multiple update types, including firmware updates, emoji/sticker package updates, voice command library updates, and large language model configuration updates.

A standard integration cycle takes 1–2 weeks to produce samples and 4–6 weeks for mass production. Customers need to provide: ① product hardware schematic (for reviewing UART/MIC/speaker interfaces) ② target wake words and command word list ③ cloud integration requirements (such as a specified large language model or IoT platform). Waytronic provides: UART serial communication protocol documentation, development and debugging tools (serial port tool + APP), and technical support covering solution design → DEMO validation → testing and acceptance → mass production.

Taking an AI electric toothbrush as an example: the original solution uses ESP32-S2 (RMB 15–20) + an external voice chip (RMB 5–8) + an external display driver (RMB 3–5), resulting in a total BOM cost of RMB 23–33. Upgrading to the WT3000A-M06 chipset solution to replace multiple discrete components can reduce the BOM by approximately 30%. At the same time, standby power consumption can be reduced from 10–15 μA to 2 μA, enabling up to 120 days of extended battery life. By replacing redundant external components with an integrated chipset solution, customers can achieve broader functionality while reducing per-unit BOM cost.

Deliverables include: ① chip/module specifications ② UART serial communication protocol documentation ③ network provisioning mini program/operation instructions ④ development and debugging tools (serial port tool + APP) ⑤ technical support. Schematics are provided, and customers can receive assistance with PCBA design. Hardware modifications are minimized by retaining the mature architecture that has already been validated for mass production. The existing MCU, motor driver, sensors, and charging circuit can all be retained, eliminating the need for new tooling and significantly reducing mass-production modification costs and implementation time.

WT2606A has passed industrial-grade certification, supporting a wide operating temperature range of -40°C to +85°C and ESD HBM protection of ±4 kV. Waytronic has served more than 30,000 customers across 80+ industry segments. For the medical and healthcare sector, including blood pressure monitors, blood glucose meters, and patient monitors, dedicated Class II medical device AI voice interaction solutions are available. These solutions support the complete healthcare workflow, including on-device offline wake-up, voice playback of measurement results, AI health interpretation, voice queries of historical data, AI health consultation, voiceprint recognition, and voice cloning.