makcedward/nlpaug

Data augmentation for NLP

What it solves

Many machine learning models require large amounts of labeled data to perform well, but creating this data manually is time-consuming. nlpaug provides a way to automatically generate synthetic training data by augmenting existing text, audio, and spectrogram data, which helps improve model performance and robustness without manual effort.

How it works

The library uses a system of Augmenter objects to modify data. It supports three primary modalities:

  • Textual: Modifies characters (simulating keyboard or OCR errors), words (using synonyms, antonyms, or embeddings like BERT, RoBERTa, and Word2Vec), and sentences (using GPT-2 or XLNet).
  • Audio: Applies signal processing techniques such as adjusting pitch, speed, volume, and injecting noise.
  • Spectrogram: Applies masking (time and frequency) and volume adjustments to visual representations of audio.

Users can use a Flow to create pipelines that apply multiple augmentation functions sequentially or randomly.

Who it’s for

It is designed for machine learning engineers and data scientists who need to expand their training datasets for NLP, audio, or speech recognition projects.

Highlights

  • Multimodal Support: Handles text, audio, and spectrograms in one library.
  • Coded for Integration: Plug-and-play compatibility with common neural network frameworks.
  • Extensive Augmentation Methods: Includes everything from simple random swaps to advanced contextual word embeddings via Transformers.
  • Pipeline Orchestration: Ability to chain multiple augmenters together via the Flow system.

Related

  • Project
    iver56/audiomentations

    A Python library for audio data augmentation that provides a wide variety of transforms to make audio deep learning models more robust for real-world use.

  • Project
    facebookresearch/AugLy

    A multimodal data augmentation library that provides over 100 transforms for audio, image, text, and video to improve AI model robustness against real-world internet data.

  • Project
    lhotse-speech/lhotse

    A Python library for flexible multimodal data preparation that optimizes the loading and manipulation of speech, audio, video, and image data for ML training.

  • Dispatch
    Hugging Face TextImage Augmentation pipeline release

    Hugging Face and Albumentations AI released a TextImage Augmentation pipeline that jointly modifies document images and their text, enabling realistic synthetic data generation and robust fine‑tuning of vision‑language models on limited document datasets.

  • Project
    QData/TextAttack

    A Python framework for generating adversarial examples, augmenting data, and training robust NLP models to identify and fix vulnerabilities in text-based AI.