Training Keyword Spotting Models on Non-IID Data with Federated Learning

TL;DR

This paper shows that high-quality keyword-spotting models can be trained on-device with federated learning, achieving performance comparable to centralized training despite non-IID data, by optimizing algorithms and using data augmentation.

Contribution

It introduces methods for effective federated training of keyword-spotting models on non-IID data, including hyperparameter tuning, SpecAugment, and teacher-student training.

Findings

SpecAugment reduces false reject rate by 56%

Federated learning achieves comparable performance to centralized models

Thorough empirical studies optimize federated training on non-IID data

Abstract

We demonstrate that a production-quality keyword-spotting model can be trained on-device using federated learning and achieve comparable false accept and false reject rates to a centrally-trained model. To overcome the algorithmic constraints associated with fitting on-device data (which are inherently non-independent and identically distributed), we conduct thorough empirical studies of optimization algorithms and hyperparameter configurations using large-scale federated simulations. To overcome resource constraints, we replace memory intensive MTR data augmentation with SpecAugment, which reduces the false reject rate by 56%. Finally, to label examples (given the zero visibility into on-device data), we explore teacher-student training.