AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Stream RAG: Instant and Accurate Spoken Dialogue Systems with Streaming Tool Usage

Siddhant Arora, Haidar Khan, Kai Sun, Xin Luna Dong, Sajal Choudhary, Seungwhan Moon, Xinyuan Zhang, Adithya Sagar, Surya Teja Appini, Kaushik Patnaik, Sanat Sharma, Shinji Watanabe and 5 more

ICML 2026 · Large language models › Retrieval and knowledge · approach: Long-context and personalized retrieval · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add web and KG tool retrieval (open book) to speech-in speech-out modelQwen-OMNI · AudioCRAG-Syntheticaccuracy %: 11.1 -> 26.3Improved
add tool retrieval (open book)Qwen-OMNI · AudioCRAG-Syntheticfirst-token latency (s): 1.34 -> 5.90Hurt
add Model-Triggered Stream RAG post-training and streaming queries over open bookQwen-OMNI · AudioCRAG-Syntheticaccuracy %: 26.3 -> 34.2Improved
add Model-Triggered Stream RAG over open bookQwen-OMNI · AudioCRAG-Humantool use latency (s): 3.37 -> 1.57Improved
serve Model-Triggered Stream RAG with vLLMQwen-OMNI · AudioCRAG-Humanmedian first-token latency (s): 3.16 -> 1.36Improved
post-train sequential RAG instead of streaming RAG (ablation)Qwen-OMNI · AudioCRAG-Syntheticaccuracy %: 34.9 vs 34.2No clear effect
post-train with streaming tool query generation instead of sequential tool integration (ablation)OpusLM · AudioCRAG-Syntheticaccuracy %: 22.4 -> 23.6No clear effect
use self-cascade (text then speech) for speech output (ablation)Qwen-OMNI open book · AudioCRAG-Syntheticaccuracy %: 26.3 -> 33.8Improved
use model-triggered instead of fixed-interval trigger (ablation)Stream RAG · AudioCRAGparallel tool threads and latency savingsImproved
use fixed-interval Stream RAG without post-training (ablation)Qwen-OMNI open book · AudioCRAGtool use latency: 10.7% reductionImproved
use Stream RAG instead of cascaded pipelineQwen-OMNI · SLUE-SQAexact match accuracy: 36.6 -> 55.8Improved
add speech output instead of text output with Stream RAGQwen-OMNI · AudioCRAG-Syntheticaccuracy %: 39.8 -> 34.2Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy