# Uzbek Automatic Speech Recognition (ASR) using the USC Corpus

This dataset card describes the Uzbek Speech Corpus (USC) and its associated Automatic Speech Recognition (ASR) system.  This repository provides the code used in the paper "[USC: An Open-Source Uzbek Speech Corpus](https://arxiv.org/abs/2107.14419)".  The ASR system is built upon ESPnet and Kaldi.  Pre-trained models are available.

**Repository:** [Link to Hugging Face Repository (replace with actual link)]

## Summary

This repository contains the code and recipes for training and using an Automatic Speech Recognition (ASR) system for the Uzbek language, utilizing the open-source Uzbek Speech Corpus (USC). The system is built upon the ESPnet framework and uses Kaldi for some functionalities.  Pre-trained Conformer models are provided.  The repository requires the prior installation of ESPnet and Kaldi.


## Authors

* Muhammadjon Musaev
* Saida Mussakhojayeva
* Ilyos Khujayorov
* Yerbolat Khassanov
* Mannon Ochilov
* Huseyin Atakan Varol


## Citation

```bibtex
@article{musayev2021usc,
  title={USC: An Open-Source Uzbek Speech Corpus},
  author={Musaev, Muhammadjon and Mussakhojayeva, Saida and Khujayorov, Ilyos and Khassanov, Yerbolat and Ochilov, Mannon and Varol, Huseyin Atakan},
  journal={arXiv preprint arXiv:2107.14419},
  year={2021}
}
```

## Related Projects and Libraries

* **ESPnet:** [https://github.com/espnet/espnet](https://github.com/espnet/espnet)
* **Kaldi:** (Implicit dependency through ESPnet)


## Pre-trained Models

Pre-trained Conformer models are available.  (Specific download location omitted as per instructions)


## Data Splits

Information on data splits is available in the original paper and dataset itself (links omitted as per instructions)


## Contact Information

(Contact information may be added here if available)