mask-Net: Learning Context Aware Invariant Features using Adversarial Forgetting (Student Abstract)

TL;DR

This paper introduces Mask-Net, a novel adversarial forgetting approach to learn invariant features for speech-to-text systems, improving generalization and reducing errors across different datasets.

Contribution

The paper proposes a new adversarial forgetting method to induce invariance in features, enhancing robustness in speech recognition models.

Findings

Achieved 2.2% absolute WER improvement on out-of-distribution data.

Achieved 1.3% absolute WER improvement on in-distribution data.

Demonstrated better generalization compared to traditional models.

Abstract

Training a robust system, e.g.,Speech to Text (STT), requires large datasets. Variability present in the dataset such as unwanted nuisances and biases are the reason for the need of large datasets to learn general representations. In this work, we propose a novel approach to induce invariance using adversarial forgetting (AF). Our initial experiments on learning invariant features such as accent on the STT task achieve better generalizations in terms of word error rate (WER) compared to the traditional models. We observe an absolute improvement of 2.2% and 1.3% on out-of-distribution and in-distribution test sets, respectively.