Acoustic Decision Support · PS-S01 Healthcare

Listening to
what the lung
is saying.

EchoAssist converts a breathing recording into a visual signal, reads its acoustic structure, and explains which part of the sound drove every classification.

Sound becomes signal becomes understanding

Begin Analysis

Live acoustic signal · SR 4 kHz

01Listen

02Clean

03Analyze

04Explain

05Confidence

Listen

Raw respiratory audio recorded at 4 kHz. Every breath cycle becomes an individual sample.

Clean

Downsampling to 4 kHz discards noise above 2 kHz. Peak normalisation equalises amplitude.

Analyze

Mel-spectrogram converts sound to a 64×157 image. ResNet18 classifies: normal, crackle, wheeze, or both.

Explain

Occlusion analysis finds the most influential second. Grad-CAM reveals exact pitch-frequency activations.

Confidence

Multi-label binary scores tell you not just what was found, but how certain the model is — and when to decline.

The Pipeline

Acoustic intelligence, step by step.

Scroll to trace the signal from raw breath to clinical insight.

Signal Exploration

Look closer.
Understand the signal.

Move your cursor over the spectrogram to focus the acoustic lens.
Salient region highlighted when a classification is available.

Product

Analyse a recording.

Drop a file here or click to select
WAV · MP3 · FLAC · OGG · up to 50 MB

Or choose a curated sample

Loading samples…

Listening towhat the lungis saying.

Acoustic intelligence, step by step.

Look closer.Understand the signal.

Analyse a recording.

Listening to
what the lung
is saying.

Look closer.
Understand the signal.