LIVE RESEARCH PROTOTYPENOT FOR CLINICAL DIAGNOSIS

03 / Dataset

A dataset shaped by listening.

The working collection organizes respiratory sounds into research-ready clips and labels, with provenance kept close to every example.

Research artifactInteractive figure
Dataset processing pipelineClick to inspect · Esc to close
Research artifactInteractive figure
Configuration parametersClick to inspect · Esc to close
Research artifactInteractive figure
Configuration numbers to knowClick to inspect · Esc to close
Dataset observatory / Respiratory Sound Database

Listen by cohort.

Explore a local, illustrative index of the Kaggle Respiratory Sound Database before jumping to the authoritative source.

SOURCE / KAGGLEView dataset ↗
920WAV files
126patients
6,898annotated cycles
5.5 haudio captured
Selected recording

PUL-0042

COPD cohort · Wheeze

Wheeze
Signal preview
Duration8.4 s
Sampling16 kHz
Acoustic markerContinuous tonal band
ANNOTATIONS / 00:0000:08

Signal preview is illustrative. Use the Playground to run a model-style prediction demo.

4target sound labels
Audioprimary modality
Openresearch posture
01

Wheeze

Continuous musical tones or high-pitched sounds.

02

Crackle

Short, discontinuous popping events.

03

Normal

Breath sounds without target abnormalities.

04

Mixed / review

Ambiguous clips retained for careful analysis.

SHWASA · RESPIRATORY ACOUSTIC INTELLIGENCEOPEN METHODS / INSPECTABLE OUTPUTS