Small-footprint models and algorithms are crucial in case of running services locally on IoT/Wearable/Consumer devices. These devices cannot offer a large memory space or computational power, and are not continuously connected to a global network.

Therefore the models have to be reduced in size, and algorithms optimised for the available local computational resources. The use-case example is to run a voice trigger together with a domain-specific ASR and NLP locally, to provide full control of a device such as a light controller.