Beyond Classification: A Cough Regression Benchmark for Respiratory Acoustic Foundation Models

2026-06-13 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

A new multi-model, multi-target cough regression benchmark evaluates five respiratory acoustic foundation models (FMs) across six health targets on three datasets. It explores their ability to predict continuous quantities like age, BMI, and disease probability from cough audio, moving beyond classification. Findings indicate MLP-small regression heads outperform baselines in 23 of 30 cases, while full MLP overfits on small data but recovers on larger sets, revealing a dataset size x head-capacity trade-off. HeAR leads within-dataset age regression on Coswara (9.12 yr MAE). HeAR and M2D+Resp achieve near-full performance at N = 50 samples, significantly more data-efficient than OPERA models requiring N = 400. Cross-dataset transfer is strongly asymmetric, with large diverse data generalizing better to small clinical populations.

Key takeaway

For machine learning engineers developing respiratory acoustic models, prioritize MLP-small regression heads for robust performance on continuous health quantity prediction. Be aware that HeAR and M2D+Resp offer high data efficiency, reaching near-full performance with N=50 samples. When considering transfer learning, focus on pretraining with large, diverse datasets to ensure generalizability to smaller clinical populations.

Key insights

Respiratory acoustic foundation models can predict continuous health quantities from cough audio using regression, moving beyond classification.

Principles

• Optimal regression head capacity depends on dataset size.
• Large, diverse datasets enable asymmetric cross-dataset transfer.

Method

The benchmark evaluates five FMs on six targets across three datasets using linear, MLP-small, and full MLP regression heads under subject-disjoint protocols.

In practice

• HeAR achieves 9.12 yr MAE for age regression on Coswara.
• HeAR and M2D+Resp reach near-full performance at N=50 samples.

Topics

• Cough Regression
• Respiratory Acoustics
• Foundation Models
• Machine Learning Benchmarks
• Transfer Learning
• Health Monitoring

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Pretrained Neural Audio Models for Asthma Detection from Voice and Speech — Pretrained neural audio models can effectively detect asthma from mobile-recorded speech, outperforming scratch training.
• QDSP: An Interpretable Structured Learning Framework for Predicting Death or Cerebral Palsy in Very Low Birth Weight Infants — QDSP is an interpretable structured learning framework for predicting outcomes in very low birth weight infants.
• Benchmarking Machine Learning Architectures for Antimicrobial Stewardship in Pediatric ICUs — Machine learning performance for pediatric antimicrobial stewardship is driven more by target design and data characteristics than model complexity.
• Towards a General Intelligence and Interface for Wearable Health Data — Large-scale pretraining on wearable data creates a foundation model for diverse health insights.
• Hybrid deep learning model for multimodal vocal and lung signal analysis in health monitoring — A hybrid deep learning model effectively integrates vocal and lung signal analysis for accurate disease detection.
• Language Models as Semantic Teachers: Post-Training Alignment for Medical Audio Understanding — Aligning audio encoders with medical LLMs via a "semantic teacher" paradigm significantly enhances clinical diagnostic performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.