Tiny-CRNN: Streaming Wakeword Detection In A Low Footprint Setting

09/29/2021

∙

by   Mohammad Omar Khursheed, et al.

∙

0

∙

share

In this work, we propose Tiny-CRNN (Tiny Convolutional Recurrent Neural Network) models applied to the problem of wakeword detection, and augment them with scaled dot product attention. We find that, compared to Convolutional Neural Network models, False Accepts in a 250k parameter budget can be reduced by 25 Tiny-CRNN architecture, and we can get up to 32 a 50k parameter budget with 75 word-level Dense Neural Network models. We discuss solutions to the challenging problem of performing inference on streaming audio with this architecture, as well as differences in start-end index errors and latency in comparison to CNN, DNN, and DNN-HMM models.