DiffATR: Diffusion-based Generative Modeling for Audio-Text Retrieval

TL;DR

DiffATR introduces a diffusion-based generative framework for audio-text retrieval, modeling joint distributions to improve out-of-domain performance and combining generative and discriminative training strategies.

Contribution

The paper proposes a novel diffusion-based generative model for ATR that captures joint distributions and enhances out-of-domain retrieval capabilities.

Findings

Outperforms existing methods on AudioCaps and Clotho datasets.

Demonstrates robustness in out-of-domain retrieval scenarios.

Combines generative and discriminative training for improved performance.

Abstract

Existing audio-text retrieval (ATR) methods are essentially discriminative models that aim to maximize the conditional likelihood, represented as p(candidates|query). Nevertheless, this methodology fails to consider the intrinsic data distribution p(query), leading to difficulties in discerning out-of-distribution data. In this work, we attempt to tackle this constraint through a generative perspective and model the relationship between audio and text as their joint probability p(candidates,query). To this end, we present a diffusion-based ATR framework (DiffATR), which models ATR as an iterative procedure that progressively generates joint distribution from noise. Throughout its training phase, DiffATR is optimized from both generative and discriminative viewpoints: the generator is refined through a generation loss, while the feature extractor benefits from a contrastive loss, thus combining the merits of both methodologies. Experiments on the AudioCaps and Clotho datasets with superior performances, verify the effectiveness of our approach. Notably, without any alterations, our DiffATR consistently exhibits strong performance in out-of-domain retrieval settings.