Keyword spotting is one of the most widespread TinyML applications. A microphone feeds short audio frames to a compact neural network that outputs the probability of a few trained keywords. Because the model is small, it runs continuously on a microcontroller or a low-power DSP at a tiny fraction of the main processor’s power budget, while that processor stays in deep sleep.
The design trades vocabulary for efficiency: the model recognizes a handful of words, not open speech, and hands off to a cloud or on-device speech recognizer once triggered. Accuracy depends heavily on noise and accents in the training data, and false triggers waste power and annoy users. The guide TinyML: machine learning on microcontrollers uses keyword spotting as a worked example.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.