CNN spectrogram
Learns local time-frequency patterns from mel representations.
02 / Models
Small models. Clear signals.
We compare complementary views of the same respiratory sound instead of hiding the decision behind a single score.
Research artifactInteractive figure
Research artifactInteractive figure
Temporal encoder
Tracks how respiratory events unfold across a recording.
Soft ensemble
Combines model views to make predictions more stable and reviewable.
“The useful model is the one a researcher can inspect, challenge, and improve.”