File size: 2,399 Bytes
22a167d
 
 
 
 
 
 
 
 
 
 
 
82ea54a
 
 
 
34cf3c3
22a167d
 
 
 
 
34cf3c3
 
22a167d
34cf3c3
22a167d
34cf3c3
22a167d
 
34cf3c3
22a167d
34cf3c3
22a167d
a69afec
34cf3c3
 
 
dd14e9c
22a167d
15f645f
22a167d
15f645f
22a167d
 
 
 
 
 
 
 
 
 
 
 
34cf3c3
22a167d
 
 
15f645f
34cf3c3
4978c70
 
 
 
 
 
22a167d
 
 
 
 
 
 
82ea54a
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
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
---
library_name: transformers
license: mit
base_model: openai/whisper-large-v3-turbo
tags:
- generated_from_trainer
metrics:
- wer
- bleu
model-index:
- name: whisper-large-v3-turbo-FLEURS-GL-EN
  results: []
datasets:
- juanjucm/FLEURS-SpeechT-GL-EN
language:
- gl
- en
---


# whisper-large-v3-turbo-FLEURS-GL-EN

This model is a fine-tuned version of [openai/whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) trained on [juanjucm/FLEURS-SpeechT-GL-EN](https://huggingface.co/datasets/juanjucm/FLEURS-SpeechT-GL-EN)
for **Galician-to-English Text to Speech Translation** task. It takes galician speech audios as input and generates the correspondant translated transcription in English.

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 BLEU score of **5.0** on the evaluation dataset.

After fine-tuning, it achieves the following results on the evaluation set:
- Loss: 1.4958
- Wer: 71.6814
- **BLEU: 18.9665**
- **ChrF++: 46.00**


The following hyperparameters were used during training:

- learning_rate: 5e-06
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- total_train_batch_size: 32
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4
- mixed_precision_training: Native AMP
- 

### Training results

We used [BLEU Score](https://en.wikipedia.org/wiki/BLEU) as our reference translation metric for selecting the best checkpoint after training.

| Training Loss | Epoch | Step | Validation Loss | Wer      | Bleu    |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:-------:|
| 4.2751        | 1.0   | 5    | 3.8850          | 76.6962  | 18.0512 |
| 2.3984        | 2.0   | 10   | 2.6965          | 97.0501  | 13.5327 |
| 1.4958        | 3.0   | 15   | 2.2308          | 71.6814  | 18.9665 |
| 1.27          | 4.0   | 20   | 2.0454          | 128.3186 | 12.2446 |


### Framework versions

- Transformers 4.45.1
- Pytorch 2.4.1+cu121
- Datasets 3.0.1
- Tokenizers 0.20.0