Audio Analysis
Audio Chunk / Slicer
Cuts each clip into fixed-length windows — overlapping if you like — one row and clip per window, the last one padded with silence to full length (or kept short, or dropped).
When to use it
Models with a fixed input size (a CNN on 5-second spectrograms, an embedding index of equal chunks) need equal pieces; overlap keeps an event at a boundary whole in one of them.
In a pipeline
Audio File InputAudio Chunk / SlicerLog-Mel Spectrogram
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Window length (seconds) | number | — | 5 |
| Overlap (seconds) | number | — | 1 |
| Last, shorter window | select | — | pad |
Example
Five-second windows, one second of overlap
- Window length (seconds)=5
- Overlap (seconds)=1
| File Name | Chunk | Start (s) | End (s) |
|---|---|---|---|
| engine.wav | 1 | 0 | 5 |
| engine.wav | 2 | 4 | 9 |
| engine.wav | 3 | 8 | 12.4 |
Worth knowing
- With padding, End (s) is where the audio really ends; the clip itself is still full length.
- Window 5 s with overlap 1 s starts a window every 4 s.