The idea was to train an audio classifier to recognize finger movement on the laptop chassis, specifically for inputting digits. It worked halfway.
- Classifying variable-length audio files with all-convolutional networks and masked global pooling, https://arxiv.org/abs/1607.02857