Reproduce “Towards Pre-training an Effective Respiratory Audio Foundation Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Daisuke Niizumi, Daiki Takeuchi, Masahiro Yasuda, Binh Thien Nguyen, Yasunori Ohishi, Noboru Harada
NTT Communication Science Laboratories' research establishes effective pre-training strategies for respiratory audio foundation models, showing that models pre-trained on large-scale general audio datasets, particularly M2D, achieve superior performance on the OPERA benchmark, with M2D reaching an average AUROC of 0.789 and further re-training on respiratory data boosting it to 0.800.
Key insights
- Pre-training on large, diverse general audio datasets (e.g., AudioSet) is more effective for respiratory audio analysis than pre-training on smaller, domain-specific datasets.
- Further pre-training general audio models on a combination of general and respiratory datasets leads to additional performance gains.
- Preserving frequency-wise information in feature aggregation is critical for respiratory audio analysis, outperforming traditional mean pooling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.