nlpaug

Why we included this project

Nlpaug generates synthetic training data for machine learning projects by altering text, audio, and spectrogram inputs, which is useful when your dataset is too small or too repetitive to train a robust model. It ships a range of augmenters covering synonym and antonym swaps, contextual word substitution with BERT-class models, back-translation, and simulated spelling and keyboard errors, so you can target the specific weakness you want to harden against. The library keeps the core API small: a single Augmenter class handles individual operations, and a Flow pipeline chains several of them sequentially or at random. That makes it straightforward to slot into an existing training loop, and the included example notebooks let teams check whether an augmentation approach works before committing to it.