SignalAutomotive
Sign in
Research example / not a diagnosis

What a real labelled engine-audio example can—and cannot—do.

This experiment turns a public BMW M54B25 dataset into four small acoustic reference groups. It demonstrates the product pipeline without pretending one car supplies universal BMW coverage.

Source vehicle
2004 BMW · BMW M54B25
Recordings
39 total · 34 used here
Training audio
About 19.1 minutes
Licence
Apache License 2.0
On-device research sandbox

Compare a recording with the four small reference groups.

Signal measures basic acoustic features in your browser, then compares them with numerical summaries derived from 34 labelled training clips. It does not upload the recording or run a component diagnosis.

Choose a short WAV, MP3, M4A or video clip.

How it is integrated

The audio becomes evidence, not an answer.

  1. 01

    Measure

    Loudness, frequency balance, zero crossings and changes over time are calculated locally.

  2. 02

    Compare

    Those measurements are compared with derived summaries of four labelled recording groups.

  3. 03

    Limit

    If a clip falls outside the tiny reference set, Signal says so instead of forcing a label.

  4. 04

    Verify later

    Real product accuracy needs more engines, independent test vehicles and confirmed workshop outcomes.

Help broaden the evidence

One engine is a start. Confirmed repair outcomes make it useful.

Contribute a properly labelled case, join the garage pilot or help fund broader dataset collection and evaluation.