---
license: mit
language:
- uz
tags:
- audio
- speech
task_categories:
- automatic-speech-recognition
---


# Uzbek Speech Corpus (USC)

**Paper:** [USC: An Open-Source Uzbek Speech Corpus and Initial Speech Recognition Experiments](https://link.springer.com/chapter/10.1007/978-3-030-87802-3_40)

**Summary:** This repository contains dataset for reproducing the results presented in the paper "USC: An Open-Source Uzbek Speech Corpus". USC is a freely available, manually checked speech corpus comprising 958 speakers and 105 hours of transcribed audio recordings. 

**Dataset Summary:**

| Feature          | Description                                      |
|-----------------|--------------------------------------------------|
| Language         | Uzbek                                           |
| Size             | 105 hours of audio                               |
| Number of Speakers | 958                                             |
| Transcription   | Manual, checked by native speakers                |
| ASR Architectures | DNN-HMM and End-to-End (E2E)                     |
| WER (Validation) | 18.1%                                           |
| WER (Test)       | 17.4%                                           |


**Github Repository:** https://github.com/IS2AI/Uzbek_ASR

**Citation:**
```bibtex
@inproceedings{musaev2021usc,
  title={USC: An open-source Uzbek speech corpus and initial speech recognition experiments},
  author={Musaev, Muhammadjon and Mussakhojayeva, Saida and Khujayorov, Ilyos and Khassanov, Yerbolat and Ochilov, Mannon and Atakan Varol, Huseyin},
  booktitle={Speech and Computer: 23rd International Conference, SPECOM 2021, St. Petersburg, Russia, September 27--30, 2021, Proceedings 23},
  pages={437--447},
  year={2021},
  organization={Springer}
}
```