Towards hate speech detection in low-resource languages: Comparing ASR to acoustic word embeddings on Wolof and Swahili

TL;DR

This paper compares automatic speech recognition and acoustic word embeddings for hate speech keyword detection in low-resource languages, finding AWE more robust in real-world scenarios with minimal training data.

Contribution

It demonstrates that acoustic word embeddings can outperform traditional ASR in low-resource, real-world hate speech detection tasks with limited labeled data.

Findings

AWE performs better in in-the-wild scenarios with minimal data

ASR outperforms AWE in controlled, same-domain experiments

AWE requires significantly less training data to achieve comparable performance

Abstract

We consider hate speech detection through keyword spotting on radio broadcasts. One approach is to build an automatic speech recognition (ASR) system for the target low-resource language. We compare this to using acoustic word embedding (AWE) models that map speech segments to a space where matching words have similar vectors. We specifically use a multilingual AWE model trained on labelled data from well-resourced languages to spot keywords in data in the unseen target language. In contrast to ASR, the AWE approach only requires a few keyword exemplars. In controlled experiments on Wolof and Swahili where training and test data are from the same domain, an ASR model trained on just five minutes of data outperforms the AWE approach. But in an in-the-wild test on Swahili radio broadcasts with actual hate speech keywords, the AWE model (using one minute of template data) is more robust, giving similar performance to an ASR system trained on 30 hours of labelled data.