Saved in:
Bibliographic Details
Main Authors: Wijngaard, Gijs, Formisano, Elia, Esposito, Michele, Dumontier, Michel
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.06815
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916834581676032
author Wijngaard, Gijs
Formisano, Elia
Esposito, Michele
Dumontier, Michel
author_facet Wijngaard, Gijs
Formisano, Elia
Esposito, Michele
Dumontier, Michel
contents Audio question answering (AQA) requires models to understand acoustic content and perform complex reasoning. Current models struggle with dataset imbalances and unstable training dynamics. This work combines curriculum learning with statistical data balancing to address these challenges. The method labels question difficulty using language models, then trains progressively from easy to hard examples. Statistical filtering removes overrepresented audio categories, and guided decoding constrains outputs to valid multiple-choice formats. Experiments on the DCASE 2025 training set and five additional public datasets show that data curation improves accuracy by 11.7% over baseline models, achieving 64.2% on the DCASE 2025 benchmark.
format Preprint
id arxiv_https___arxiv_org_abs_2507_06815
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Data-Balanced Curriculum Learning for Audio Question Answering
Wijngaard, Gijs
Formisano, Elia
Esposito, Michele
Dumontier, Michel
Sound
Audio and Speech Processing
Audio question answering (AQA) requires models to understand acoustic content and perform complex reasoning. Current models struggle with dataset imbalances and unstable training dynamics. This work combines curriculum learning with statistical data balancing to address these challenges. The method labels question difficulty using language models, then trains progressively from easy to hard examples. Statistical filtering removes overrepresented audio categories, and guided decoding constrains outputs to valid multiple-choice formats. Experiments on the DCASE 2025 training set and five additional public datasets show that data curation improves accuracy by 11.7% over baseline models, achieving 64.2% on the DCASE 2025 benchmark.
title Data-Balanced Curriculum Learning for Audio Question Answering
topic Sound
Audio and Speech Processing
url https://arxiv.org/abs/2507.06815