makcedward/nlpaug

Features

• Augmentation Pipelines - Provides a pipeline orchestrator to chain multiple augmentation functions sequentially or randomly for synthetic dataset generation.
• Machine Learning Data Augmentation - Acts as a comprehensive library for creating diverse variations of text, audio, and spectrogram training data.
• Training Data Augmentation - Provides techniques for diversifying audio datasets through signal manipulation and noise injection to improve model robustness.
• Back-Translation Data Augmentation - Implements synthetic text generation by translating samples to a foreign language and back to the original source.
• Synthetic Dataset Generators - Functions as a synthetic dataset generator to increase the size and variety of NLP training data.
• Rule-Based Text Transformations - Simulates typos, spelling errors, and word substitutions using predefined dictionaries and character-level mapping rules.
• Stochastic Augmentation Pipelines - Executes a series of augmentation functions sequentially or randomly to create diverse synthetic datasets.
• Textual Variation Generators - Produces new sentence variations using back-translation and contextual embeddings to broaden training data.
• Embedding-Based Substitutions - Replaces or substitutes words in text using pre-trained embeddings and contextual models.
• LM-Based Text Generation - Generates synthetic text by substituting, inserting, or deleting characters and words via language models.
• Rule-Based Alterations - Alters content using synonym replacement, keyboard mistake simulation, and spelling errors.
• Sentence Generation - Creates new sentence variations using abstractive summarization and contextual generation.
• Textual Data Synthesizers - Creates variations of text through synonym replacement, back-translation, and contextual word embeddings.
• Semantic Substitutions - Provides semantic word substitution using pre-trained embedding models to maintain meaning while diversifying training text.
• Machine Learning Pipelines - Implements a pipeline orchestrator to chain multiple data transformation steps for generating synthetic training datasets.
• Audio Signal Augmentors - Provides a set of tools for modifying audio files via noise injection, pitch shifting, and time stretching.
• Language Model Integrations - Integrates external language models and embeddings to guide the generation of contextual synthetic text data.
• Text Data Augmentation - Utilizes established language models and word embeddings to inform the generation of synthetic textual data.
• Textual Noise Simulations - Simulates realistic typos and keyboard mistakes in text datasets to train more resilient NLP models.
• NLP Pipeline Orchestrators - Provides a system for composing augmentation functions into sequential data flows for synthetic dataset creation.
• Spectrogram Generation - Produces synthetic images from audio data to provide diverse visual representations for model training.
• Frequency Masking - Implements time and frequency masking on audio spectrograms to force machine learning models to learn more robust features.
• Spectrogram Masking - Transforms audio spectrograms using time and frequency masking to improve speech recognition robustness.
• Acoustic Signal Modification - Modifies audio speed, pitch, and normalization to increase acoustic variety in synthetic audio datasets.
• Noise and Time Shifting - Creates synthetic audio samples by applying noise injection and time shifting to improve model robustness.
• Spectrogram Masking Tools - Transforms audio spectrograms using time and frequency masking to improve speech recognition robustness.
• Data Augmentation - Data augmentation library for English NLP.

Star history