Peer-reviewed research
SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
Daniel S. Park et al., Proceedings of Interspeech 2019, 2019
Read the originalRead on October 17, 2026
- What it is
- A Google Brain paper that proposes SpecAugment, a data augmentation method applied to the filter bank features a speech network reads. It warps the features and masks blocks of frequency channels and time steps, and it reports results on LibriSpeech and Switchboard.
- How much weight it carries
- Peer-reviewed at Interspeech 2019 and widely reused. The reported results are for a 2019 Listen, Attend and Spell model on read and telephone speech, so later systems differ. Augmentation changes how a model is trained, not how it runs.
Facts taken from this source
- SpecAugment is a data augmentation method that is applied directly to a speech network's input features, the filter bank coefficients. It warps the features and masks blocks of frequency channels and blocks of time steps.
“SpecAugment is applied directly to the feature inputs of a neural network (i.e., filter bank coefficients). The augmentation policy consists of warping the features, masking blocks of frequency channels, and masking blocks of time steps.” (Abstract)