Timestamps and weak labels for training an engine-configuration audio classifier (v-twin vs.
inline-4 vs. flat-6, etc.) from short audio windows. This dataset does not contain audio.
Each row points at a public YouTube video id plus a (start_sec, end_sec) window; you fetch
and slice the audio yourself (see Reconstructing audio below).
The source audio was collected by searching YouTube (via… See the full description on the dataset page:
https://huggingface.co/datasets/joakes90/Auto_Engine_Classification.
