Towards Effective Negation Modeling in Joint Audio-Text Models for Music
This work addresses a specific limitation in music retrieval systems for users needing accurate semantic distinctions, but it is incremental as it builds on existing CLAP models and datasets.
The paper tackled the problem of joint audio-text models struggling with semantic negation in music retrieval, such as distinguishing 'with vocals' from 'without vocals', by training CLAP models with text augmentation and a contrastive loss, resulting in improved negation handling while largely preserving retrieval performance.
Joint audio-text models are widely used for music retrieval, yet they struggle with semantic phenomena such as negation. Negation is fundamental for distinguishing the absence (or presence) of musical elements (e.g., "with vocals" vs. "without vocals"), but current systems fail to represent this reliably. In this work, we investigate and mitigate this limitation by training CLAP models from scratch on the Million Song Dataset with LP-MusicCaps-MSD captions. We introduce negation through text augmentation and a dissimilarity-based contrastive loss, designed to explicitly separate original and negated captions in the joint embedding space. To evaluate progress, we propose two protocols that frame negation modeling as retrieval and binary classification tasks. Experiments demonstrate that both methods, individually and combined, improve negation handling while largely preserving retrieval performance.