Omni-Embed-Audio: Leveraging Multimodal LLMs for Robust Audio-Text Retrieval
Omni-Embed-Audio (OEA) is a retrieval encoder using multimodal LLMs with native audio understanding to improve audio-text retrieval beyond standard caption queries. It introduced User-Intent Queries (UIQs) and hard negative mining to test robustness. OEA significantly outperforms baselines in text-to-text retrieval and demonstrates superior ability to distinguish between acoustically similar distractors.