|
--- |
|
license: apache-2.0 |
|
language: |
|
- lv |
|
pipeline_tag: automatic-speech-recognition |
|
base_model: |
|
- openai/whisper-large-v3 |
|
--- |
|
|
|
# General-purpose Latvian ASR model |
|
|
|
This is a fine-tuned [whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) model for Latvian, trained by [AiLab.lv](https://ailab.lv) using two general-purpose speech datasets: the Latvian part of [Common Voice 19.0](https://commonvoice.mozilla.org/lv/datasets), and the latest version of the Latvian broadcast dataset [LATE-Media](https://korpuss.lv/id/LATE-mediji). |
|
|
|
This version of the model supersedes the previous [whisper-large-v3-lv-late-cv17](https://huggingface.co/AiLab-IMCS-UL/whisper-large-v3-lv-late-cv17) model. |
|
|
|
We also provide 4-bit, 5-bit and 8-bit quantized versions of the model in the GGML format for the use with [whisper.cpp](https://github.com/ggerganov/whisper.cpp), as well as an 8-bit quantized version for the use with [CTranslate2](https://github.com/OpenNMT/CTranslate2). |
|
|
|
## Training |
|
|
|
Fine-tuning was done using the Hugging Face Transformers library with a modified [seq2seq script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/speech-recognition#sequence-to-sequence). |
|
|
|
| Training data | Hours | |
|
|:---|---:| |
|
| Latvian Common Voice 19.0 train set (the [VW split](https://analyzer.cv-toolbox.web.tr)) | 212.6 | |
|
| LATE-Media 2.0 train set | 69.8 | |
|
| Total | 282.4 | |
|
|
|
## Evaluation |
|
|
|
| Testing data | WER | CER | |
|
|:---|---:|---:| |
|
| Latvian Common Voice 19.0 test set (VW) - formatted | 4.8 | 1.6 | |
|
| Latvian Common Voice 19.0 test set (VW) - normalized | 3.2 | 1.0 | |
|
| LATE-Media 1.0 test set - formatted | 19.2 | 7.6 | |
|
| LATE-Media 1.0 test set - normalized | 12.8 | 5.3 | |
|
|
|
The Latvian CV 19.0 test set is available [here](https://analyzer.cv-toolbox.web.tr). |
|
The LATE-Media 1.0 test set is available [here](http://hdl.handle.net/20.500.12574/99). |
|
|
|
## Citation |
|
|
|
Please cite this paper if you use this model in your research: |
|
|
|
```bibtex |
|
@inproceedings{dargis-etal-2024-balsutalka-lv, |
|
author = {Dargis, Roberts and Znotins, Arturs and Auzina, Ilze and Saulite, Baiba and Reinsone, Sanita and Dejus, Raivis and Klavinska, Antra and Gruzitis, Normunds}, |
|
title = {{BalsuTalka.lv - Boosting the Common Voice Corpus for Low-Resource Languages}}, |
|
booktitle = {Proceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING)}, |
|
publisher = {ELRA and ICCL}, |
|
year = {2024}, |
|
pages = {2080--2085}, |
|
url = {https://aclanthology.org/2024.lrec-main.187} |
|
} |
|
``` |
|
|
|
## Acknowledgements |
|
|
|
This work was supported by the EU Recovery and Resilience Facility project [Language Technology Initiative](https://www.vti.lu.lv) (2.3.1.1.i.0/1/22/I/CFLA/002) in synergy with the State Research Programme project [LATE](https://www.digitalhumanities.lv/projekti/vpp-late/) (VPP-LETONIKA-2021/1-0006). |
