ViToSA: Audio-Based Toxic Spans Detection on Vietnamese Speech Utterances

TL;DR

This paper presents ViToSA, a novel Vietnamese audio dataset for toxic speech detection, and a pipeline combining ASR and toxic span detection that improves accuracy and establishes a new benchmark for low-resource language content moderation.

Contribution

Introduction of ViToSA, the first Vietnamese toxic speech dataset, and a combined ASR and toxic span detection pipeline that enhances toxic speech identification accuracy.

Findings

Fine-tuning ASR models on ViToSA reduces transcription errors.

Text-based toxic span detection models outperform existing baselines.

Establishment of a new benchmark for Vietnamese audio toxic speech detection.

Abstract

Toxic speech on online platforms is a growing concern, impacting user experience and online safety. While text-based toxicity detection is well-studied, audio-based approaches remain underexplored, especially for low-resource languages like Vietnamese. This paper introduces ViToSA (Vietnamese Toxic Spans Audio), the first dataset for toxic spans detection in Vietnamese speech, comprising 11,000 audio samples (25 hours) with accurate human-annotated transcripts. We propose a pipeline that combines ASR and toxic spans detection for fine-grained identification of toxic content. Our experiments show that fine-tuning ASR models on ViToSA significantly reduces WER when transcribing toxic speech, while the text-based toxic spans detection (TSD) models outperform existing baselines. These findings establish a novel benchmark for Vietnamese audio-based toxic spans detection, paving the way for future research in speech content moderation.