---
type: "post"
id: "new-open-dataset-synth-dopaco-releases-8-800-91cde0"
slug: "new-open-dataset-synth-dopaco-releases-8-800-91cde0"
title: "New open dataset Synth-DoPaCo releases 8,800 synthetic doctor-patient conversations with 1,329 hours of audio."
description: "Synth-DoPaCo makes a long-context audio task more inspectable by packaging synthetic clinical conversations, audio, and reference notes together."
retrieval_nugget: "Synth-DoPaCo makes a long-context audio task more inspectable by packaging synthetic clinical conversations, audio, and reference notes together."
published_at: "2026-10-18"
updated_at: "2026-10-18"
record_date: "2026-10-18"
date_kind: "published_at"
topics: ["evals","multimodal"]
source_urls: ["https://arxiv.org/abs/2604.06138"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:new-open-dataset-synth-dopaco-releases-8-800-91cde0"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:bba92138cb2ec48e2dfed2fb1a0f0f8cb368623ac5837987d883fe4829cdcbea","reviewed_at":"2026-10-18T02:01:46.885Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"A dataset designed for both training and controlled evaluation can expose failure modes that short-context benchmarks miss.","observed_facts":[{"text":"The paper reports 8,800 synthetic first-visit doctor-patient conversations and about 1.3 thousand hours of associated audio with reference SOAP notes.","source_urls":["https://arxiv.org/abs/2604.06138"]},{"text":"The pipeline uses persona-driven dialogue generation, multi-speaker audio with overlap and pauses, room acoustics and sound events, and LLM-based reference notes.","source_urls":["https://arxiv.org/abs/2604.06138"]},{"text":"The published version corrected a bug and reports per-claim hallucination rates of 32% for end-to-end systems and 22–24% for cascaded systems.","source_urls":["https://arxiv.org/abs/2604.06138"]}],"mechanism":"The pipeline controls scenario and acoustic features while pairing synthetic audio with reference SOAP notes, creating a shared artifact for training and controlled evaluation.","why_now":"Long-context audio reasoning remains underserved, while open-ended summaries are difficult to evaluate automatically.","implications":["Researchers can use the release to test long-form summarization and compare system designs against shared references.","Synthetic benchmark performance does not demonstrate clinical validity or transfer to real patient recordings."],"evidence_boundary":"The supplied paper excerpt describes a synthetic first-visit dataset and corrected comparisons; it does not establish performance on real clinical recordings or deployment readiness.","watch_conditions":["Whether independent evaluations reproduce the comparison on real-world clinical audio and inspect dataset and reference coverage and errors."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied for this specific dataset release; this starts a distinct analytical record focused on its controlled-evaluation design and corrected results."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/new-open-dataset-synth-dopaco-releases-8-800-91cde0/","markdown":"https://newruntime.com/posts/new-open-dataset-synth-dopaco-releases-8-800-91cde0.md","json":"https://newruntime.com/posts/new-open-dataset-synth-dopaco-releases-8-800-91cde0.json"}
related_patterns: ["scientific-agents-need-artifact-evidence"]
---

# New open dataset Synth-DoPaCo releases 8,800 synthetic doctor-patient conversations with 1,329 hours of audio.

## Retrieval answer

Synth-DoPaCo makes a long-context audio task more inspectable by packaging synthetic clinical conversations, audio, and reference notes together.

Synth-DoPaCo makes a long-context audio task more inspectable by packaging synthetic clinical conversations, audio, and reference notes together.

Thesis: A dataset designed for both training and controlled evaluation can expose failure modes that short-context benchmarks miss. The paper reports 8,800 synthetic first-visit doctor-patient conversations and about 1.3 thousand hours of associated audio with reference SOAP notes. The pipeline uses persona-driven dialogue generation, multi-speaker audio with overlap and pauses, room acoustics and sound events, and LLM-based reference notes.

Mechanism: The pipeline controls scenario and acoustic features while pairing synthetic audio with reference SOAP notes, creating a shared artifact for training and controlled evaluation. Why now: Long-context audio reasoning remains underserved, while open-ended summaries are difficult to evaluate automatically. The published version corrects a bug in earlier results; the authors report per-claim hallucination rates of 32% for end-to-end systems and 22–24% for cascaded systems, describing performance as comparable.

First implication: Researchers can use the release to test long-form summarization and compare system designs against shared references. Evidence boundary: The supplied paper excerpt describes a synthetic first-visit dataset and corrected comparisons; it does not establish performance on real clinical recordings or deployment readiness. First watch condition: Whether independent evaluations reproduce the comparison on real-world clinical audio and inspect dataset and reference coverage and errors. This is a useful experimental resource, not proof of deployment readiness.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Evals](https://newruntime.com/topics/evals.md): Explore the evals topic hub.
- topic hub: [Multimodal](https://newruntime.com/topics/multimodal.md): Explore the multimodal topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)