TensorFlow is powering everything from data centers to edge devices, across industries from finance to advanced healthcare. And now, with TensorFlow 2.0 and the evolving ecosystem of tools and libraries, it is doing it all so much easier. At this year's TensorFlow World, Google and Arm are distributing various Adafruit PyBadges with TensorFlow Lite Micro (TF-Lµ) pre-installed. These dev boards demonstrate how TF-Lµ can perform inference, offline, on low-cost, low-power Arm microcontrollers. They are running the TF-Lµ Micro Speech example and set to recognize the words "yes" and "no". It is easy to retrain this board using any of the Micro Speech vocabulary. Use this guide to building TensorFlow Lite and training example models all in a Docker image.
The Adafruit PyBadge makes flashing new models as easy as dragging and dropping new models on to the board, as a USB Mass Storage device, similar to this "Cat Dog" speech recognition example. Training can take several hours on modern laptops.
Your conference board includes a battery and a microphone. All that you have to do is switch it on, press the "A" button and speak "yes" or "no". You might recognize the response animation and sound. We invite you to retrain your boards, tinker with the responses, and investigate using other examples like the Magic Wand gesture recognition with the built-in accelerometer. Feel free to ask questions and share your projects!
Find out more about TensorFlow world:
[CTAToken URL = "https://conferences.oreilly.com/tensorflow/tf-ca" target="_blank" text="Find out more" class ="green"]
Should you completely erase your bootloader or badge demo, here are the files and instructions you'll need to get the demo back to normal.