MFCC Temporal Manipulations — User Guide
Praat script.
What this does
Praat script.
Process Description: This script analyzes a selected Sound object using MFCCs (Mel-Frequency Cepstral Coefficients) and uses the resulting spectral information to drive time-domain resynthesis parameters (Pitch Tier and Duration Tier) via a Manipulation object. MFCCs are commonly used in speech recognition to represent the vocal tract shape. By manipulating sound based on the evolution of these coefficients over time, we can create complex, spectral-aware sound transformations.
Quick start
- In Praat, select a Sound object.
- Run script… →
MFCC Temporal Manipulations.praat. - Set parameters and click OK.
- Output object appears (see Outputs).
Parameters (form fields)
| Name (GUI) | Type | Default |
|---|---|---|
| Number_of_coefficients | positive | 12 |
| Window_length_(s) | positive | 0.015 |
| Time_step_(s) | positive | 0.005 |
| First_filter_frequency_(Hz) | positive | 100 |
| Distance_between_filters_(Hz) | positive | 100 |
| Maximum_frequency_(Hz) | real | 0 |
| Complexity_threshold | positive | 0.5 |
| Max_stretch_factor | positive | 2.0 |
| Min_stretch_factor | positive | 0.5 |
| Freeze_duration_(s) | positive | 0.2 |
| Similarity_threshold | positive | 0.3 |
| Scramble_window_(frames) | positive | 10 |
| Keep_intermediate_objects | boolean | 0 |
Outputs
- Object name: output
- Sample rate: unknown
- Normalization: No automatic normalization