File size: 2,761 Bytes
ebfb239
 
 
 
 
91847e3
 
ebfb239
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
---
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).

## 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).