SSZTAN9 november   2016


  1.   1
  2.   2
    1.     3
    2.     Additional resources:

Lorin Netsch

In our connected world we are seeing more electronic devices that understand speech. Smartphones, tablets and laptops include apps like Siri or Cortana that let help you search for answers, control the electronics around you and more. While these apps are impressive, they also take a lot of processing power and memory. So it is not a surprising misconception that microcontrollers (MCUs) are just too small to recognize speech.


It is true that the low-power and small size goals of an MCU may not allow it to understand everything someone could say but for small low-power embedded applications, all you may need is to recognize a few well-defined phrases. For example, like “heat up my coffee” or “turn the lights off.” Recently we have demonstrated this feature on our low-power MSP432™ MCU.

We have released a Voice Detection Plugin for SimpleLink MSP432 SDK that can enable an MSP432 MCU-based application to recognize your own personal speech phrases. It recognizes up to 11 phrases while ignoring other speech. You do have to tell the recognizer your phrases by saying them a few times but once that’s done, it’s ready to go.

The Voice Detection Plugin has features that you would expect such as:

  • You can change the phrases any time you want
  • It can be instructed to respond to only a few of the phrases
  • And of course, you can delete a phrase from your repertoire if you wish

The Voice Detection Plugin comes with some easy to use header files and user and API guides to get you up and running quickly. The Plugin also includes an example demo program targeting the MSP432 MCU LaunchPad™ development kit, Audio BoosterPack™ plug-in  module, and the Kentec LCD BoosterPack  kit.

MSP432 Microcontroller Speech Recognition Technology video

The demo app uses the 14-bit analog-to-digital converter (ADC14) integrated in the MSP432 MCU to gather speech, and the LCD to display menus. The menus allow you to run recognizer features. You can choose to:

  • Say a phrase that you want the recognizer to remember. It will create a model of that speech and store it in flash memory (a task called enrollment).
  • Say an enrolled phrase again. The recognizer will use it to make a better model for improved performance (a task called update).
  • Delete a model that has been enrolled
  • Run recognition

So, what are your ideas for MCU-based apps and devices that could be voice enabled? I’d be interested in hearing. But now I think I’ll take a break to “Heat up my coffee.”