Audio Analysis
MFCC Extractor
Mel-frequency cepstral coefficients: a compact, low-dimensional description of each clip's timbre — summarised as the mean and spread of each coefficient, or one row per frame.
When to use it
Lightweight classifiers (MLDoo, scikit-learn, a decision tree) work well on MFCC summaries: one numeric row per clip, ready to train on.
In a pipeline
Audio File InputResamplerMFCC ExtractorExport CSV
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Coefficients | number | — | 13 |
| Mel bands | number | — | 40 |
| Window (ms) | number | — | 25 |
| Hop (ms) | number | — | 10 |
| Add deltas (rate of change) | checkbox | — | false |
| Results | select | — | summary |
Example
13 coefficients, summarised
- Coefficients=13
- Results=One row per clip: mean and spread of each coefficient
| File Name | MFCC 1 Mean | MFCC 1 Std | MFCC 2 Mean |
|---|---|---|---|
| dog-07.wav | -312.4 | 41.2 | 88.9 |
Worth knowing
- Add deltas to capture how the sound changes, not just what it sounds like.
- Computed like librosa's defaults (dB mel bands, orthonormal DCT), so values are comparable with models trained in Python.