neuspell

Why we included this project

NeuSpell is a research toolkit that exposes ten neural spelling correctors through a single Python API, so you can swap between CNN-LSTM, SC-LSTM, and BERT-based checkers without rewriting the rest of your pipeline. Its models are trained on real misspellings and synthetic noise, and they handle naturally occurring typos better than older tools like Aspell. The package also includes utilities for generating synthetic misspellings and fine-tuning checkers on your own domain data, plus a web demo for manual testing. You can measure correction accuracy as you go. If your pipeline handles text where typos are common, like search queries or chat messages, the pretrained checkpoints and the option to adapt them to custom vocabularies are what you will use most.