# Forgive or forget: Understanding the context of hate in audio retrieval systems

arXiv: https://arxiv.org/abs/2606.05857 · HTML: https://mlsanity.com/paper/2606.05857/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 8 of its 16 references that are in the map.
- First posted on arXiv 4 Jun 2026.

## Builds on (most cited first, 8 of 8)

- [Automated Hate Speech Detection and the Problem of Offensive Language](https://mlsanity.com/paper/1703.04009/) (2017)
- [Noise2Noise: Learning Image Restoration without Clean Data](https://mlsanity.com/paper/1803.04189/) (2018)
- [Clotho: An Audio Captioning Dataset](https://mlsanity.com/paper/1910.09387/) (2019)
- [WavCaps: A ChatGPT-Assisted Weakly-Labelled Audio Captioning Dataset for Audio-Language Multimodal Research](https://mlsanity.com/paper/2303.17395/) (2023)
- [Audio-Text Models Do Not Yet Leverage Natural Language](https://mlsanity.com/paper/2303.10667/) (2023)
- [MuTox: Universal MUltilingual Audio-based TOXicity Dataset and Zero-shot Detector](https://mlsanity.com/paper/2401.05060/) (2024)
- [NOMAD: Unsupervised Learning of Perceptual Embeddings for Speech Enhancement and Non-matching Reference Audio Quality Assessment](https://mlsanity.com/paper/2309.16284/) (2023)
- [Dissecting Temporal Understanding in Text-to-Audio Retrieval](https://mlsanity.com/paper/2409.00851/) (2024)

## Abstract (arXiv)

Handling toxic retrieval in text-to-audio systems is challenging due to contextual dependencies. Existing strategies (e.g., rephrasing, summarization) risk altering intent or omitting details. We propose a post hoc causal debiasing framework with a sentiment-controlled mediator to preserve semantic relevance while suppressing harmful speech. Our approach is model-agnostic and integrates seamlessly with existing retrieval pipelines. We introduce two variants: Forgive, which re-ranks and filters toxic audio via logit adjustment, and Forget, which generates counterfactual toxic prompts to mitigate harmful retrievals. Experiments show consistent toxicity reduction with minimal loss in retrieval accuracy, improving both safety and reliability.