This week’s EYE ON NPI (video) looks at a new spin on a familiar favorite! We’ve chatted before about the IMX RT10xx series of chips from NXP. These ‘cross-over’ chips are Cortex M7’s running at 500 or 600 MHz – that means you get the relative simplicity of a Cortex microcontroller with the speed and power of a microprocessor. Heck, the RT106x’s ~3000 CoreMark @ 600 MHz makes it faster for some computations when you compare it to the Raspberry Pi Zero’ ~2000 CoreMark – most of that due to the Linux operating system overhead. This chip family recently made its way into CircuitPython as we were super-excited to see such a powerful and low cost chip.
OK but back to this week’s EYE ON NPI which is a sister-chip, the iMX RT106L – an interesting twist to the RT1060 – this chip is licensed to use with a wake-word + voice-activation software stack. It seems like the hardware is the same, and share the same datasheets and reference manuals. So it’s more about purchasing hardware that has a licensing cost baked in – this reminds us of when we would purchase MP3 chips that had a 50 cent licensing fee built in.
So, why is this interesting? Well, you’ve seen the proliferation of voice assistants like Siri, Google Voice and Alexa – they all use a a microcomputer, with a very powerful processor, to record audio and send it to the cloud for processing. Having to build a full computer with operating system to do voice control makes the BoM cost go up – you need a few MB of RAM, plenty of flash, a complex PMIC…what if we could do all of the processing on a really fast microcontroller instead? Like…a 600 MHz M7? Yes!
NXP’s MCU-based solution for local voice control leverages the i.MX RT106L crossover MCU, enabling developers to quickly and easily add local voice commands to their products. The ultra-small form-factor, production-ready hardware design comes with fully integrated software that runs on FreeRTOS for quick, out-of-the-box evaluation and proof-of-concept development. This turnkey solution minimizes time-to-market, risk, and development effort, enabling OEMs to easily add voice to their smart home and smart appliance products without the need for Wi-Fi and cloud connectivity while also addressing the privacy concerns of many consumers.
There’s a few different products in this family. First is the 106L – the L for “Local”. That’s internet-less voice command inference. Good for maybe a dozen words. There’s a collection of words built-in – for more words you’ll have to train up the engine, which is a non-trivial task. The Local is great for simple device control and no WiFi is required so it’s very low power and low cost.
Next is the similar 106A, the A is “Alexa”, as you may guess it has Amazon Alexa software ready to go to create custom Alexa-connected devices. You’ll need to use/add WiFi for the cloud backend communication. Digi-Key has a great tutorial/article about the Alexa series and the nitty-gritty detail of how it works.
Lastly is the 106F where F is for “Face” – yes facial recognition! This version connects up to a camera and can be used to recognize faces.
These are all interesting machine learning demos many of us have seen, and its interesting to see NXP integrate them as software ‘add-on’s for a very powerful and capable chip. Note that the RT1060 doesn’t have hardware ML acceleration or co-processor, so you’ll be relying on the speed and power of the microcontroller itself. We think in the next few years we will see similar “AI ready” chips that are even less expensive, thanks to a neural-net accelerator.
Meanwhile, if you want to explore these powerful chips, you can check out the entire NXP iMX RT series chip collection here on Digi-Key.
And pick up the iMX RT106L eval kit here on Digi-Key (part 568-SLN-LOCAL-IOT-ND) so you can start playing with local voice recognition: order today and you’ll have it tomorrow!