Two options of augmentation:
Option 1: Preferred – during augmentation, the dataset side increases. The generated new samples are part of the user’s data sets.
Option 2: Not sure if it is needed for timeseries. Looks like lower priority. Augmentation is happening during training only. The augmented samples are not temporary for that training session. The user has control over the type of augmentation.

Examples of sound data augmentation:

• Adding different noise for audio dataset: White, brown, etc.
• Change the sampling rate ( from 100 kHz to 10 kHz).
• Change of volume

For images, There are many augmentations to prioritize.