nlpaug 0.0.3

Natural language processing augmentation library for deep neural networks

nlpaug

This python library helps you with augmenting nlp for your machine learning projects. Visit this introduction to understand about Data Augmentation in NLP

Feature

• Provide both character and word level augmentations which include:
  • Character Augmentation: OCR, QWERTY(Keyboard Distance), Random Behavior
  • Word Augmentation:
    • Random Behavior: RandomWord
    • Synonym: WordNet
    • Word Embeddings: word2vec, GloVe, fasttext,
    • Language Models: BERT
  • Speech Recognition Augmentation:
    • Spectrogram: Frequency Masking, Time Masking
    • Audio: Noise, Pitch, Shift, Speed
• Flow orchestration is supported. Flow includes:
  • Sequential: Apply data augmentations one by one
  • Sometimes: Apply some augmentations randomly

Example

• How to use character and word augmentation
• How to create custom augmentation
• How to use spectrogram augmentation for speech recognition
• How to use audio augmentation for speech recognition

Frequency Masking

Time Masking

Installation

The library supports python 3.5+ in linux and window platform.

To install the library:

pip install nlpaug

Download word2vec or GloVe files if you use Word2VecAug or GloVeAug:

• word2vec(GoogleNews-vectors-negative300)
• GloVe(glove.6B.50d)
• fasttext(wiki-news-300d-1M.vec.zip)

Recent Changes

0.0.3 May 23, 2019: Added Speed, Noise, Shift and Pitch augmenters for Audio

0.0.2 Apr 30, 2019: Added Frequency Masking and Time Masking for Speech Recognition (Spectrogram). Added librosa library dependency for converting wav to spectrogram.

0.0.1 Mar 20, 2019: Project initialization

Test

Word2vec and GloVe models are used in word insertion and substitution. Those model files are necessary in order to run test case. You have to add ".env" file in root directory and the content should be
	- MODEL_DIR={MODEL FILE PATH}

Folder structure of model should be
	-- root directory
		- glove.6B.50d.txt
		- GoogleNews-vectors-negative300.bin
		- wiki-news-300d-1M.vec