Audio Analysis
Resampler
Converts every clip to one sample rate with a band-limited (windowed-sinc) filter, so lowering the rate never folds high frequencies back as noise.
When to use it
Put it before any model: Whisper, Wav2Vec 2.0 and the AudioSet classifier expect 16 kHz, and mixed-rate inputs make features incomparable.
In a pipeline
Audio File InputResamplerLog-Mel Spectrogram
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Sample rate | select | — | 16000 |
| Sample rate (Hz) | number | — | 32000 |
Example
A CD-quality file down to 16 kHz
- Sample rate=16 kHz (speech ML)
| File Name | Sample Rate |
|---|---|
| song.flac | 44100 |
| File Name | Sample Rate |
|---|---|
| song.flac | 16000 |
Worth knowing
- The model nodes resample on their own as well; resample first when the same clips feed several nodes, or for the DSP nodes.
- File Name keeps the original name; Audio Path points at the converted clip in the run's working folder.