Small-footprint-Keyword-Spotting work on Google Speech Commands Dataset [link]
Kaldi implementation using DNN-HMM model
End-to-end model using TDNN with Domain Adversarial Training
Run models on android sysyem to test latency and accuracy, implementing by PyTorch Mobile [link] and Vosk-API [link]