Clinical audio models show different effects from pairing metadata accurately
Correspondence Is Not Transfer: A Pairing-Controlled Audit of Clinical Audio-Metadata Alignment
Sound
Summary
When computer programs study health sounds, they often use extra information about the person who made the sound. This additional info can be tricky because it includes many things, not just illness details. The authors tested what happens when they keep or shuffle the matches between sounds and this info. They found that matching sounds exactly to the right metadata helps find some personal traits better, like sex, but it's still unclear if it helps detect diseases consistently. So, linking sounds and data well affects some parts of analysis but not all in the same way.
What this means in practice
- •For clinical data teams: Improve quality checks by separating effects of metadata pairing from label information in audio-based health datasets.
- •For machine learning engineers: Evaluate and refine contrastive training methods to better understand metadata influence on disease prediction models from audio.
Authors
Liu He, Xi Wang, Jiahong Yuan, Yuanchao Li
Abstract
Clinical audio models increasingly align recordings with participant metadata or metadata-derived text. This supervision is useful but ambiguous: the paired target can contain symptoms, demographics, missingness, and cohort structure, not just disease information. We study this ambiguity by making the audio-metadata pairing itself the intervention. Our Pairing-Controlled Transfer Audit trains matched contrastive trajectories under correct pairing, global shuffling, disease-label-preserving shuffling, and label-and-recorded-sex-preserving shuffling. Each endpoint is then frozen and tested with three readouts: profile correspondence, attribute decodability, and matched disease prediction. On UKCOVID, correct improves retrieval and sex decodability over label-only shuffling, showing that the pairing intervention worked. The sex-probe contrast is largely absorbed by the sex-preserving control, while smaller retrieval differences remain. Disease-transfer contrasts, however, remain uncertain. After UKCOVID discovery fixed the pipeline, CODA TB, Cambridge COVID-19 Sounds, and Coswara were rerun as a locked external panel with the same arms, backbones, matching, readout, calibration, primary contrast, and primary metric. The external panel preserves the descriptive separation without proving a universal null. Successful profile correspondence and improved disease transfer are therefore different empirical claims.