Talking to Microcontrollers

The newly released Picovoice Shepherd makes voice recognition on MCUs point-and-click easy.

Nick Bild
3 years agoMachine Learning & AI
(📷: Picovoice)

If you have ever worked with an enterprise grade development toolchain — full of acronyms and convoluted workflows — on a hobby project where you just wanted to get busy building your project, but instead found yourself mired in unintelligible documentation, then you may appreciate a new tool released by Picovoice. Picovoice Shepherd offers a simple way to add voice recognition—similar to that seen in Amazon’s Alexa — to microcontroller-powered devices.

The no-code platform builds machine learning models that run entirely on-device. No Internet connectivity is needed, and the latency and privacy concerns inherent in cloud-based solutions do not apply.

The web-based tool, Picovoice Console, is used to create and train models in a graphical environment. Within this tool, the Porcupine Wake Word Engine is used to train a continuously-listening model to recognize a phrase that will activate the voice recognition system. Rather than collecting lots of samples of the wake word being spoken, and designing and training a model with the help of that data, one needs only to type the wake word phrase. The console hides the details, but in the background uses transfer learning to generate an accurate and robust model.

The next step in the Console is the Rhino Speech-to-Intent Engine, which infers intent from spoken commands. This enables you to map voice commands to intended actions. There is a helpful, built-in feature that allows testing of the intent engine as it is being designed. Once satisfied with the results, a button click will build the model — again, there is no data collection, model design, or parameter specification.

Once a model has been created in the Picovoice Console, it can then be debugged, tested, and transferred to a microcontroller from the Shepherd tool. A number of popular Arm Cortex-M microcontrollers from NXP and STMicroelectronics are currently supported.

Picovoice Shepherd is free for personal and research use. Take a look at the documentation to get started building your own voice recognition model.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles