Synthetic Doctor-Patient Conversations with Audio and SOAP Notes

BeTraC 2026 is a synthetic dataset containing 7,600 doctor-patient dialog sessions with corresponding audio, transcripts, and SOAP note summaries. The dataset was created by BeTraC and released in April 2026. It is structured into training and development splits, with the training set comprising 7,200 dialogs.

Use Cases

• Train automatic speech recognition models on the .opus audio files paired with dialog transcripts.
• Develop multi-modal models that generate SOAP note summaries from audio or transcript dialog metadata.
• Build intent classification or named entity recognition models using the structured dialog metadata fields.
• Create synthetic data augmentation pipelines for medical conversation AI using the 7,200 training dialogs.

Strengths

• Contains 7,600 total synthetic doctor-patient conversation dialogs.
• Provides multi-modal data per sample: audio, transcript, dialog metadata, and SOAP note summary.
• Training split is substantial with 7,200 dialogs across 9 shards totaling 8.7 GB.

Limitations

• Data is entirely synthetic, which may not fully capture the complexity and variability of real clinical conversations.
• No information is provided on the diversity of medical conditions, patient demographics, or speaker accents covered.
• The dataset uses the WebDataset format, which may require specific tooling for efficient loading and processing.

Provenance

Source

BeTraC

Collection Method

Synthetically generated doctor-patient conversations.

Freshness

Last updated on 2026-04-03.