---
license: cc-by-4.0
language:
- en
task_categories:
- token-classification
size_categories:
- 10M<n<100M
---

This is the training dataset used to train the first version of the SpellMapper model.
It consists of 33'547'074 training examples.
Actually, [this checkpoint](https://huggingface.co/bene-ges/spellmapper_asr_customization_en) was trained only 1/3 of the full dataset.

Paper: [SpellMapper: A non-autoregressive neural spellchecker for ASR customization with candidate retrieval based on n-gram mappings](https://arxiv.org/abs/2306.02317)

## Citation

```bibtex
    @misc{antonova2023spellmapper,
      title={SpellMapper: A non-autoregressive neural spellchecker for ASR customization with candidate retrieval based on n-gram mappings}, 
      author={Alexandra Antonova and Evelina Bakhturina and Boris Ginsburg},
      year={2023},
      eprint={2306.02317},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
    }
```