EfficientNetB0 + BiLSTM + Attention
Binary COPD screening. Output: COPD probability.
87.73%held-out accuracy02 / Models
Two models. Two questions.
The report compares complementary research tasks, not a single leaderboard: binary COPD screening and four-class respiratory sound-event recognition.
Custom four-stage residual CNN
Four-class respiratory sound-event recognition. Output: Sound-event class.
92.56%official ICBHI scoreResearch artifactInteractive figure
Research artifactInteractive figure
Shared input contract
Both tracks begin with patient-aware ICBHI audio preparation and normalized time-frequency representations.
Different outputs
COPD-EFF produces a binary screening probability; copd-mod2 recognizes four acoustic events.
Complementary evidence
The report presents the tracks as complementary capabilities, not a direct ranking.
“The useful model is the one a researcher can inspect, challenge, and improve.”