|
--- |
|
library_name: transformers |
|
license: mit |
|
base_model: openai/whisper-large-v3-turbo |
|
tags: |
|
- generated_from_trainer |
|
metrics: |
|
- wer |
|
model-index: |
|
- name: whisper-large-v3-turbo-OpenSLR-GL |
|
results: [] |
|
datasets: |
|
- juanjucm/OpenSLR-SpeechT-GL-EN |
|
language: |
|
- gl |
|
--- |
|
|
|
# whisper-large-v3-turbo-OpenSLR-GL |
|
|
|
This model is a fine-tuned version of [openai/whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) trained on [juanjucm/OpenSLR-SpeechT-GL-EN](https://huggingface.co/datasets/juanjucm/OpenSLR-SpeechT-GL-EN) for **Galician Text to Speech** task. It takes galician speech audios as input and generates the correspondant transcription. |
|
|
|
This Automatic Speech Recognition model, was developed to be the first stage of a Speech Translation cascade system for transcribing and translating Galician audios into English texts. After this first STT step, this [Galician-to-English MT model](https://huggingface.co/juanjucm/nllb-200-distilled-600M-OpenSLR-GL-EN) can be applied over the generated Galician transcriptions to get English text translations. |
|
|
|
The motivation behind this work is to increase the visibility of the Galician language, making it more accessible for non-Galician speakers to understand and engage with Galician audio content. |
|
|
|
This model was developed during a 3-week Speech Translation workshop organised by [Yasmin Moslem](https://huggingface.co/ymoslem). |
|
|
|
### Performance and training details |
|
|
|
Baseline model achieved a WER score of **20.1** on the evaluation dataset. |
|
|
|
After fine-tuning, it achieves the following results on the evaluation set: |
|
- Loss: 0.1613 |
|
- **WER: 10.6845** |
|
|
|
The following hyperparameters were used during training: |
|
|
|
- learning_rate: 3e-06 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 2 |
|
- gradient_accumulation_steps: 2 |
|
- total_train_batch_size: 64 |
|
- total_eval_batch_size: 16 |
|
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments |
|
- lr_scheduler_type: linear |
|
- num_epochs: 10 |
|
|
|
### Training results |
|
|
|
We used [WER (Word Error Rate)](https://en.wikipedia.org/wiki/Word_error_rate) as our reference transcription metric for selecting the best checkpoint after training. |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Wer | |
|
|:-------------:|:-----:|:----:|:---------------:|:-------:| |
|
| 0.2739 | 1.0 | 75 | 0.1898 | 11.4023 | |
|
| 0.1841 | 2.0 | 150 | 0.1819 | 10.3673 | |
|
| 0.0542 | 3.0 | 225 | 0.1919 | 10.6177 | |
|
| 0.0399 | 4.0 | 300 | 0.1934 | 11.1352 | |
|
| 0.0264 | 5.0 | 375 | 0.2042 | 11.2688 | |
|
| 0.0143 | 6.0 | 450 | 0.2075 | 10.3840 | |
|
| 0.0056 | 7.0 | 525 | 0.2198 | 10.8347 | |
|
| 0.0063 | 8.0 | 600 | 0.2217 | 10.9683 | |
|
| 0.0037 | 9.0 | 675 | 0.2258 | 10.5509 | |
|
| 0.0042 | 10.0 | 750 | 0.2278 | 10.6845 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.47.1 |
|
- Pytorch 2.4.1+cu121 |
|
- Datasets 3.2.0 |
|
- Tokenizers 0.21.0 |
