{"paper":{"title":"Omni-Embed-Audio: Leveraging Multimodal LLMs for Robust Audio-Text Retrieval","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Multimodal LLM audio encoders match state-of-the-art retrieval while excelling at complex user queries and hard negatives.","cross_cats":["cs.CL"],"primary_cat":"cs.SD","authors_text":"Du-Seong Chang, HaeJun Yoo, Insung Lee, Myoung-Wan Koo, Yongseop Shin","submitted_at":"2026-04-20T14:50:33Z","abstract_excerpt":"Audio-text retrieval systems based on Contrastive Language-Audio Pretraining (CLAP) achieve strong performance on traditional benchmarks; however, these benchmarks rely on caption-style queries that differ substantially from real-world search behavior, limiting their assessment of practical retrieval robustness. We present Omni-Embed-Audio (OEA), a retrieval-oriented encoder leveraging multimodal LLMs with native audio understanding. To systematically evaluate robustness beyond caption-style queries, we introduce User-Intent Queries (UIQs) - five formulations reflecting natural search behavior"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"OEA achieves comparable text-to-audio retrieval performance to state-of-the-art M2D-CLAP, while demonstrating clear advantages in two critical areas: (1) dominant text-to-text retrieval (+22% relative improvement), and (2) substantially superior hard negative discrimination (+4.3%p HNSR@10, +34.7% relative TFR@10), revealing that LLM backbones provide superior semantic understanding of complex queries.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the five User-Intent Query formulations accurately capture real-world search behavior and that observed gains are caused by the multimodal LLM backbone rather than differences in training data, scale, or other unstated factors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Omni-Embed-Audio uses multimodal LLMs to match CLAP on standard audio retrieval while improving text-to-text retrieval by 22% relative and hard negative discrimination by 4.3 points HNSR@10 on user-intent queries.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Multimodal LLM audio encoders match state-of-the-art retrieval while excelling at complex user queries and hard negatives.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0f5fb61915f7b1ccebf1023f0a762097dfc8a791cd14010c149064ea35599d0c"},"source":{"id":"2604.18360","kind":"arxiv","version":2},"verdict":{"id":"9b406d60-952b-437f-b121-3548b4c2c03b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T03:19:17.745809Z","strongest_claim":"OEA achieves comparable text-to-audio retrieval performance to state-of-the-art M2D-CLAP, while demonstrating clear advantages in two critical areas: (1) dominant text-to-text retrieval (+22% relative improvement), and (2) substantially superior hard negative discrimination (+4.3%p HNSR@10, +34.7% relative TFR@10), revealing that LLM backbones provide superior semantic understanding of complex queries.","one_line_summary":"Omni-Embed-Audio uses multimodal LLMs to match CLAP on standard audio retrieval while improving text-to-text retrieval by 22% relative and hard negative discrimination by 4.3 points HNSR@10 on user-intent queries.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the five User-Intent Query formulations accurately capture real-world search behavior and that observed gains are caused by the multimodal LLM backbone rather than differences in training data, scale, or other unstated factors.","pith_extraction_headline":"Multimodal LLM audio encoders match state-of-the-art retrieval while excelling at complex user queries and hard negatives."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.18360/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-20T04:08:51.595166Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"a3929a5b747ec6cde5e8f566a2d8a84b496d4c2e00fa82b0820ff3819c7eaccc"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}