用合成医患对话生成长音频摘要数据
Generating Synthetic Doctor-Patient Conversations for Long-form Audio Summarization
研究人员提出一套合成数据生成流程，面向长上下文音频推理，以首次就诊医患对话和 SOAP 病历生成为任务实例，全部基于开放权重模型构建。该流程包含人格驱动对话生成、带重叠/停顿建模与房间声学及声音事件的多说话人音频合成、LLM 生成参考 SOAP 病历三个阶段，共发布 8,800 段合成对话、1.3k 小时音频及参考病历。
Authors:Yanis Labrak, David Grünert, Séverin Baroudi, Jiyun Chun, Pawel Cyrta, Sergio Burdisso, Ahmed Hassoon, David Liu, Adam Rothschild, Reed Van Deusen, Petr Motlicek, Andrew Perrault, Ricard Marxer, Thomas Schaaf
Abstract:Long-context audio reasoning is underserved in both training data and evaluation. Existing benchmarks target short-context tasks, and the open-ended generation tasks most relevant to long-context reasoning pose well-known challenges for automatic evaluation. We propose a synthetic data generation pipeline designed to serve both as a training resource and as a controlled evaluation environment, and instantiate it for first-visit doctor-patient conversations with SOAP note generation as the task. The pipeline has three stages, persona-driven dialogue generation, multi-speaker audio synthesis with overlap/pause modeling, room acoustics, and sound events, and LLM-based reference SOAP note production, built entirely on open-weight models. We release 8,800 synthetic conversations with 1.3k hours of corresponding audio and reference notes. Evaluating current open-weight systems, we find that cascaded approaches still substantially outperform end-to-end models.
| Comments: | Published at Interspeech 2026. v2 is the published version; results were corrected after a bug fix: end-to-end and cascaded systems now perform comparably (per-claim hallucination 32% vs. 22-24%) |
| Subjects: | Sound (cs.SD); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2604.06138 [cs.SD] |
| (or arXiv:2604.06138v2 [cs.SD] for this version) | |
| https://doi.org/10.48550/arXiv.2604.06138 arXiv-issued DOI via DataCite |
|
| Journal reference: | Proc. Interspeech 2026, 4384-4389 |
| Related DOI: | https://doi.org/10.21437/Interspeech.2026-2901
DOI(s) linking to related resources |
Submission history
From: Yanis Labrak [view email]
[v1]
Tue, 7 Apr 2026 17:45:07 UTC (662 KB)
[v2]
Sun, 11 Oct 2026 09:25:04 UTC (639 KB)
来源：arXiv：cs.AI · arxiv.org