Microprocessor board suggestions for Audio focused TinyML / AI

[mzvitzf]

Hi All,

I’m trying to hunt down the best board for my new project. Basically I need the following functionality either on the board or via add-on boards. I can see a few possibilities but I haven’t had much experience so I thought it best to ask. Thankyou!

Requirements

• TinyML (processed onboard) 0.5 and above Tops
• Generate and encode sinewave between 18kHz - 21kHz
• Microphone input
• Class d Speaker out (=>3w preferably)
• Low power
• Ethernet (RJ45)

Not a requirement but would be great

• POE
• Wifi
• Low budget

[Julian E.]

Hello @mzvitzf ,

Sorry for the late reply.

Seeing your requirements, all I can think of is the u85 IoT: B-U585I-IOT02A - Discovery kit for IoT node with STM32U5 series - STMicroelectronics

But it has not ethernet port and not the right audio codex.

This board is what we use for all AI audio event sound detection use case, so it is compatible with model zoo deployment for audio event detection use case as well. You can take a look here: https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/audio_event_detection/deployment

You can also take a look at the STM32U575I-EV but it is somewhat pricy and does not have a RJ45, a audio out and is not low power...

Have a good day,

Julian

[riyaz8501]

What a great query! A few standout microprocessor boards for TinyML/AI projects with an emphasis on audio are: The Arduino Nano 33 BLE Sense is small and has a built-in microphone, making it ideal for finding keywords and easy audio classification. With TinyML libraries like TensorFlow Lite for Microcontrollers, the Raspberry Pi Pico with an external microphone is both affordable and adaptable. Seeed XIAO ESP32S3 Sense – The ESP32's AI capabilities, small form factor, and onboard microphone make it an excellent choice for low-power audio tasks. The Syntiant NDP101/120 is ideal for always-on voice interfaces and is designed specifically for ultra-low power audio AI at the edge. Your power budget, model complexity, and use case (voice commands versus ambient sound classification, for example) all play a role.