# Few-Shot Contrastive Adaptation for Audio Abuse Detection in Low-Resource Indic Languages Source: [Few-Shot Contrastive Adaptation for Audio Abuse Detection in Low-Resource Indic Languages]() Collection index: [llms.txt]() Data updated: 2026-08-04T05:00:15.721Z Authors: Aditya Narayan Sankaran, Reza Farahbakhsh, Noel Crespi Published: 2026-04-10T08:23:03.000Z ## Abstract Abusive and hateful speech is increasingly spoken rather than written, surfacing in voice notes, calls, and short-form videos. Most detection systems still transcribe speech to text before classifying it, but transcription is unreliable for languages lacking strong speech recognisers, and it discards the tone and emotion that often carry the abuse itself. This paper examines whether abusive speech can instead be detected directly from audio, using CLAP, a model that learns a shared representation of sound and language, evaluated across ten Indic languages in the ADIMA dataset. A lightweight classifier trained on CLAP's existing audio representations, without adapting the model itself, comes within one to three points of a fully supervised system, and far outperforms prompting with no labelled examples at all. Further adaptation with a handful of labelled examples per language yields little extra benefit, varying unpredictably across languages. CLAP-based audio representations thus already offer a strong, inexpensive foundation for detecting abusive speech across languages, lowering the labelled data needed in practice. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Abusive Speech Detection on ADIMA Punjabi 1.0 (test)](): Abusive Speech Detection; ADIMA Punjabi 1.0 (test); Macro-F1: 83.65. - [Abusive Speech Detection on ADIMA Bhojpuri 1.0 (test)](): Abusive Speech Detection; ADIMA Bhojpuri 1.0 (test); Macro-F1: 71.48. - [Abusive Speech Detection on ADIMA Gujarati 1.0 (test)](): Abusive Speech Detection; ADIMA Gujarati 1.0 (test); Macro-F1: 75.52. - [Abusive Speech Detection on ADIMA Haryanvi 1.0 (test)](): Abusive Speech Detection; ADIMA Haryanvi 1.0 (test); Macro-F1: 80.23. - [Abusive Speech Detection on ADIMA Malayalam 1.0 (test)](): Abusive Speech Detection; ADIMA Malayalam 1.0 (test); Macro F1: 78.18. - [Abusive Speech Detection on ADIMA Odia 1.0 (test)](): Abusive Speech Detection; ADIMA Odia 1.0 (test); Macro-F1: 79.67. - [Abusive Speech Detection on ADIMA Bengali 1.0 (test)](): Abusive Speech Detection; ADIMA Bengali 1.0 (test); Macro-F1: 76.34. - [Abusive Speech Detection on ADIMA Hindi 1.0 (test)](): Abusive Speech Detection; ADIMA Hindi 1.0 (test); Macro-F1: 78.04. - [Abusive Speech Detection on ADIMA Kannada 1.0 (test)](): Abusive Speech Detection; ADIMA Kannada 1.0 (test); Macro-F1: 76.67. - [Abusive Speech Detection on ADIMA Tamil 1.0 (test)](): Abusive Speech Detection; ADIMA Tamil 1.0 (test); Macro-F1: 73.12. - [Audio Intent Classification on ADIMA](): Audio Intent Classification; ADIMA; Accuracy (Bengali): 76.34. ## Original sources and implementation [Original paper]() [PDF]()