---
license: mit
language:
- uk
pretty_name: Ukrainian subset of the YODAS dataset
---

YODAS dataset has both human-annotated and machine-generated transcriptions. 

Human-annotated transcriptions are stored in the `uk_000.parquet`. Transcriptions were artificially shortened which wasn't suitable for LLM pretraining. The shortening is reversed in the stiched version `uk_000_stitched.parquet`.

## Data source

https://huggingface.co/datasets/espnet/yodas

## Considerations for Using the Data
### Social Impact

This dataset was created to support Ukrainian language AI development and improve language technology accessibility for Ukrainian speakers.

### Recommendations

You can use this dataset for the following purposes:

* Self-supervised pretraining