Federated Learning for Keyword Spotting
We propose a practical approach based on federated learning to solve out-of-domain issues with continuously running embedded speech-based models such as wake word detectors. We conduct an extensive empirical study of the federated averaging algorithm for the "Hey Snips" wake word based on crowdsourced data on two distinct tasks: learning from scratch and language adaptation. We also reformulate the global averaging step of the federated averaging algorithm as a gradient update step, applying per-coordinate adaptive learning rate strategies such as Adam in place for standard weighted model averaging. We then empirically demonstrate that using adaptive averaging strategies highly reduces the number of communication rounds required to reach a target performance.
READ FULL TEXT