MuTox: Universal MUltilingual Audio-based TOXicity Dataset and Zero-shot Detector

TL;DR

The MuTox audio-based toxicity classifier is introduced, which enables zero-shot toxicity detection across a wide range of languages and outperforms existing text-based trainable classifiers by more than 1% AUC, while expanding the language coverage more than tenfold.

Abstract

Research in toxicity detection in natural language processing for the speech modality (audio-based) is quite limited, particularly for languages other than English. To address these limitations and lay the groundwork for truly multilingual audio-based toxicity detection, we introduce MuTox, the first highly multilingual audio-based dataset with toxicity labels. The dataset comprises 20,000 audio utterances for English and Spanish, and 4,000 for the other 19 languages. To demonstrate the quality of this dataset, we trained the MuTox audio-based toxicity classifier, which enables zero-shot toxicity detection across a wide range of languages. This classifier outperforms existing text-based trainable classifiers by more than 1% AUC, while expanding the language coverage more than tenfold. When compared to a wordlist-based classifier that covers a similar number of languages, MuTox improves precision and recall by approximately 2.5 times. This significant improvement underscores the potential of MuTox in advancing the field of audio-based toxicity detection.

Figures (6)

• Figure 1: Amount of toxicity per toxic category proposed in this paper.
• Figure 2: Toxicity selection distribution per language (x-axis) and for each method: detoxify, etox and their intersection.
• Figure 3: MuTox toxicity classifier diagram
• Figure 4: Recall per toxicity category at fixed precision of $max(\textsc{etox}\xspace, 0.3)$.
• Figure 5: Percentage (y-axis) of toxicity in the audio speech dataset in English (top) and in Spanish (bottom) per toxicity quantile (x-axis) in the text toxicity classification.