On the Role of Speech Data in Reducing Toxicity Detection Bias
Not on the map. Why?
Abstract (arXiv)
Text toxicity detection systems exhibit significant biases, producing disproportionate rates of false positives on samples mentioning demographic groups. But what about toxicity detection in speech? To investigate the extent to which text-based biases are mitigated by speech-based systems, we produce a set of high-quality group annotations for the multilingual MuTox dataset, and then leverage these annotations to systematically compare speech- and text-based toxicity classifiers. Our findings indicate that access to speech data during inference supports reduced bias against group mentions, particularly for ambiguous and disagreement-inducing samples. Our results also suggest that improving classifiers, rather than transcription pipelines, is more helpful for reducing group bias. We publicly release our annotations and provide recommendations for future toxicity dataset construction.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 29 of its 43 references that are in the map. First posted on arXiv 12 Nov 2024.
Builds on
29 of 43 references.
References · 29
- 2020wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representationscited by 3,905
- 2021WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processingcited by 1,657
- 2020The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memescited by 490
- 2023Scaling Speech Technology to 1,000+ Languagescited by 411
- 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 404
- 2024Aya Model: An Instruction Finetuned Open-Access Multilingual Language Modelcited by 295
- 2021Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detectioncited by 264
- 2019Racial Bias in Hate Speech and Abusive Language Detection Datasetscited by 209
- 2018Reducing Gender Bias in Abusive Language Detectioncited by 152
- 2023SONAR: Sentence-Level Multimodal and Language-Agnostic Representationscited by 108
- 2020Toxicity Detection: Does Context Really Matter?cited by 93
- 2022Is Your Toxicity My Toxicity? Exploring the Impact of Rater Identity on Toxicity Annotationcited by 69
17 more
- 2022Handling Bias in Toxic Speech Detection: A Surveycited by 61
- 2021Quantifying Bias in Automatic Speech Recognitioncited by 50
- 2024Decoding Biases: Automated Methods and LLM Judges for Gender Bias Detection in Language Modelscited by 33
- 2022Toxicity in Multilingual Machine Translation at Scalecited by 26
- 2019Gender Representation in French Broadcast Corpora and Its Impact on ASR Performancecited by 23
- 2021Don't speak too fast: The impact of data bias on self-supervised speech modelscited by 20
- 2024NLP Systems That Can't Tell Use from Mention Censor Counterspeech, but Teaching the Distinction Helpscited by 20
- 2020Evaluating Gender Bias in Speech Translationcited by 19
- 2022A Study of Gender Impact in Self-supervised Models for Speech-to-Text Systemscited by 15
- 2024MuTox: Universal MUltilingual Audio-based TOXicity Dataset and Zero-shot Detectorcited by 15
- 2021DeToxy: A Large-Scale Multimodal Dataset for Toxicity Classification in Spoken Utterancescited by 14
- 2022Detecting Unintended Social Bias in Toxic Language Datasetscited by 12
- 2022ADIMA: Abuse Detection In Multilingual Audiocited by 11
- 2021Perspectives on Machine Learning from Psychology's Reproducibility Crisiscited by 9
- 2021Toxicity Detection can be Sensitive to the Conversational Contextcited by 6
- 2024Enhancing Multilingual Voice Toxicity Detection with Speech-Text Alignmentcited by 5
- 2024A Comprehensive View of the Biases of Toxicity and Sentiment Analysis Methods Towards Utterances with African American English Expressionscited by 4
Built on by
Citing papers per year: 2026: 2.
2 papers.