Published April 7, 2026 | Version v1
Preprint Open

Generating Synthetic Doctor-Patient Conversations for Long-form Audio Summarization

Authors/Creators

  • 1. ROR icon Idiap Research Institute
  • 2. EDMO icon University of Zurich
  • 3. ROR icon Laboratoire d'Informatique et Systèmes
  • 4. EDMO icon Ohio State University
  • 5. Metamedia Technologies
  • 6. ROR icon Johns Hopkins University
  • 7. EDMO icon Colorado School of Mines
  • 8. solventum
  • 9. ROR icon University of Pittsburgh
  • 10. ROR icon Brno University of Technology
  • 11. ROR icon The Ohio State University
  • 12. ROR icon Pompeu Fabra University
  • 13. ROR icon Université de Toulon

Description

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.

 

Files

Interspeech_2026__Generating_Synthetic_Doctor_Patient_Conversations_for_Long_form_Audio_Summarization-5.pdf

Files (240.2 kB)

Additional details

Dates

Submitted
2026-03-04
Submitted for review at Interspeech 2026