Respiratory Sound Classification

The field studies how sounds such as wheezes, crackles, and coughs can be detected and classified to support screening and clinical assessment. Methods range from spectrogram-based CNNs and temporal models to self-supervised audio encoders, audio-text alignment, ensembles, and weakly supervised learning that identifies informative time segments. Researchers commonly evaluate on benchmarks such as ICBHI and SPRSound, while also exploring patient-level prediction, multimodal clinical context, and wearable sensors. A central challenge is that strong benchmark results often fail to transfer across patients, devices, and sites, and recognizing an acoustic event does not necessarily yield a reliable disease diagnosis.