Spectrogram Augmenter

Configure and create a spectrogram data augmentation. You can also store it in a datastore for machine learning with automated batching.

Configure and create a spectrogram data augmentation. You can also store it in a datastore for machine learning with automated batching. This implements two classes:

• spectogramDataAugmenter: configure the augmentation for the spectograms. Similar to imageDataAugmenter for images.
• augmentedSpectrogramDatastore: transform batches to augment spectrogram data. Similar to augmentedImageDatastore for images.

The techniques implented are:

• Time warp
• Frequency masking
• Time masking
• Salt and pepper noise
• White noise
• Time shift

The first three filters are from "SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition" by Google Brain and its authors Daniel S. Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D. Cubuk & Quoc V. Le.

This work is published under the CC BY-NC-SA 4.0 DEED license.