Research map / Large language models / Language, safety and interpretability
Hearing More with Less: Multi-Modal Retrieval-and-Selection Augmented Conversational LLM-Based ASR
Bingshen Mu, Hexin Liu, Hongfei Xue, Kun Wei, Lei Xie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add current-utterance hypothesis as input to LLM-ASR (ablation) | LLM-ASR · MLC-SLM test | MER (%): 11.04 -> 9.89 | Improved |
| add top-1 text-retrieved historical context (ablation) | LLM-ASR + Hyp. · MLC-SLM test | MER (%): 9.89 -> 9.23 | Improved |
| add top-1 speech-retrieved historical context (ablation) | LLM-ASR + Hyp. · MLC-SLM test | MER (%): 9.89 -> 9.41 | Improved |
| combine top-1 speech and top-1 text retrieved contexts without selection (ablation) | MARS · MLC-SLM test | MER (%): 9.23 -> 9.34 | Hurt |
| add multi-modal selection with near-ideal ranking (ablation) | MARS · MLC-SLM test | MER (%): 9.34 -> 8.96 | Improved |
| add two-pass decoding (ablation) | MARS · MLC-SLM test | MER (%): 8.96 -> 8.35 | Improved |
| increase number of preceding utterances used as context (ablation) | LLM-ASR · MLC-SLM test | MER (%): 9.74 (1 context) -> 13.49 (5 contexts) | Hurt |
| use retrieved context instead of the same number of preceding contexts (ablation) | LLM-ASR · MLC-SLM test | MER (%): 9.74 (Context{1}) -> 9.41 (speech Top-1) / 9.23 (text Top-1) | Improved |
| use near-ideal ranking instead of summing speech and text similarities (ablation) | MARS selection · MLC-SLM test | MER (%): 9.18 -> 8.96 | Improved |
| use all 2K retrieved contexts instead of selecting the best one (ablation) | MARS · MLC-SLM test | MER (%): 8.96 -> 11.24 (2xTop-3) | Hurt |
| use ground-truth preceding contexts in prior methods | Bi-context / Seewo · MLC-SLM dev | MER (%): Seewo 15.48 -> 14.30; Bi-context 14.87 -> 13.16 | Improved |
Closest papers by meaning
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.89)
- Retrieval meets Long Context Large Language Models (ICLR 2024 · similarity 0.86)
- Paralinguistics-Aware Speech-Empowered Large Language Models for Natural Conversation (NeurIPS 2024 · similarity 0.85)
- SD-Eval: A Benchmark Dataset for Spoken Dialogue Understanding Beyond Words (NeurIPS 2024 · similarity 0.85)
- F2RVLM: Boosting Fine-grained Fragment Retrieval for Multi-Modal Long-form Dialogue with Vision Language Model (AAAI 2026 · similarity 0.85)
- UniAudio 1.5: Large Language Model-Driven Audio Codec is A Few-Shot Audio Task Learner (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
