The ESC10 environmental sound dataset.
| Clip Duration | Format | Original Clips Count | Categories | Augmentation Count | Augmented Clips Count |
|---|---|---|---|---|---|
| 5 secs | .wav (originally .ogg) | 400 | 10 | 12 | 5200 |
A 5-seconds file may contain events shorter than 5 seconds, accordingly the authors of the dataset padded all files to unify the 5 seconds length for all files.
This folder contains:
- Scripts required to prepare an augmented version of the ESC10 dataset for the MCLNN processing.
- Pretrained weights and indices for the 5-fold cross-validation in addition to the standardization parameters.
The following are the steps involved in preparing the augmented ESC10 dataset:
- Follow the steps in ESC10-for-MCLNN to download and preprocess the original ESC10 dataset.
- Apply the controlled deformations for each clip using the scripts provided here.
The preparation scripts require the following packages to be installed beforehand:
- Rubber Band v1.8.1 An audio time-stretching and pitch-shifting library and utility program
- numpy 1.11.2+mkl
- librosa 0.4.0
- h5py 2.6.0
- muda 0.2.0
- Download the dataset using the ESC10_download script, make sure the files of each category are in a separate folder. If you prefer to download the dataset directly, make sure the files are ordered following the esc10aug_8pitch_4stretch_storage_ordering file.
- Position the scripts of the Preparation Scripts directory in the downloaded dataset parent directory and execute them in order following the "id_XX" index in the file name after applying any necessary configuration.
- Configure the spectrogram transformation within the Dataset Transformer and generate the MCLNN-Ready hdf5 for the dataset.
- Generate the indices for the folds using the Index Generator script.