Meta-Learning for improving rare word recognition in end-to-end ASR

TL;DR

This paper introduces a novel meta-learning approach to generate speech embeddings and adapt meta-learning methods for keyword spotting, significantly enhancing rare word recognition in end-to-end ASR systems.

Contribution

It presents new methods for meaningful speech embedding generation, modifies meta-learning approaches for continuous signal keyword spotting, and combines these into an end-to-end ASR system.

Findings

Speech embeddings perform well in few-shot settings.

Meta-learning modifications enable effective continuous signal spotting.

Word error rate improves by up to 5%.

Abstract

We propose a new method of generating meaningful embeddings for speech, changes to four commonly used meta learning approaches to enable them to perform keyword spotting in continuous signals and an approach of combining their outcomes into an end-to-end automatic speech recognition system to improve rare word recognition. We verify the functionality of each of our three contributions in two experiments exploring their performance for different amounts of classes (N-way) and examples per class (k-shot) in a few-shot setting. We find that the speech embeddings work well and the changes to the meta learning approaches also clearly enable them to perform continuous signal spotting. Despite the interface between keyword spotting and speech recognition being very simple, we are able to consistently improve word error rate by up to 5%.