Data augmentation is very standard for annotated image datasets for tasks like image labelling. Images are flipped, rotated, pixelated and so on, to add more training data and make the system robust to noise and not overfit on irrelevant features.

How are for example speech recognition training datasets pre-processed? Are they augmented with background noise or other mutations like image datasets are? Has there been any work in this direction?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.