In addition to our standard form-factor boards, Teensy is available as a member of SparkFun’s MicroMod range of M.2 mix-and-match processors. Maker Evan Rust selected this board to pair with SparkFun’s MicroMod Machine Learning Carrier for a simple yet powerful keyword recognition project.

Using publicly available datasets, Evan uploaded clips of half a dozen keywords to Edge Impulse, along with “noise” samples, and trained a Keras neural network with 92.7% recognition accuracy. Combining the resultant Arduino library generated by Edge Impulse with Teensy’s Audio Library, then downsampling from 44.1KHz to 16KHz in order to match the dataset’s samples, the inference results are displayed on an ILI9341-driven TFT in real time as a bar graph of sorts reflecting the model’s confidence. A detailed walk-through of how to recreate the project can be found on Hackster.