Audio Noise Classifier

How Audio Noise Classifier works

Press Load model once: your browser downloads YAMNet audio classifier (MediaPipe) (11 MB) from this site and keeps it in its cache. Then add your file and press Classify audio. The model runs on this device, so nothing you enter is sent to a server. Until the model is loaded, or if your device cannot run it, the tool falls back to a simple keyword check and labels the result as the fallback.

YAMNet audio classifier (MediaPipe) listens in windows of just under a second and scores each one against the 521 AudioSet sound classes, from speech and music to hum, hiss, sirens and dogs. The file is decoded in your browser and resampled to 16 kHz mono first.

The main sound is the class with the highest average score across the clip. The timeline shows the top sound in each window, merging runs of the same sound. Only the first 60 seconds are analysed.

Worked example

The model's output for the built-in sample, recorded on a desktop browser. Your run may word things slightly differently.

Synthesised for this site: 2.5 s of hiss, 2.5 s of 60 Hz mains hum, 2.5 s of 1 kHz beeps.

Main sound

Music

Also heard

Pink noise 19% · Noise 18% · Telephone 17% · Busy signal 16%

Speech

Not detected

Timeline

0.0-1.9 s: Pink noise 1.9-2.8 s: Noise 2.8-4.7 s: Music 4.7-5.6 s: Telephone 5.6-6.6 s: Beep, bleep 6.6-7.5 s: Busy signal

The model

The files are served from this site, versioned so a cached copy is never stale. Nothing is fetched from a third party while you use the tool.

Limitations

• Pure tones and steady electrical hum are often labelled as music, because the model learned from everyday recordings. Use the timeline to see where each sound starts and stops.
• It names sounds; it does not transcribe words.
• Results describe the input you gave, not the real-world situation behind it. Review them yourself before acting.

Good uses

• Podcast cleanup
• Support diagnostics
• Device QA
• Dataset labeling