File size: 2,949 Bytes
21e3286
 
 
 
 
 
 
 
 
 
 
1a79db7
180ac67
1a79db7
 
 
21e3286
 
180ac67
21e3286
180ac67
a4d7ac8
 
 
 
 
 
 
 
67901b5
a4d7ac8
40c0754
21e3286
a4d7ac8
f73eaa9
21e3286
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b9314d5
a4d7ac8
21e3286
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1a79db7
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
76
77
78
---
library_name: transformers
license: mit
base_model: openai/whisper-large-v3-turbo
tags:
- generated_from_trainer
metrics:
- bleu
model-index:
- name: whisper-large-v3-turbo-gl-en
  results: []
datasets:
- juanjucm/OpenHQ-SpeechT-GL-EN
language:
- gl
- en
---

# whisper-large-v3-turbo-OpenHQ-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/OpenHQ-SpeechT-GL-EN](https://huggingface.co/datasets/juanjucm/OpenHQ-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 **3.38** on the evaluation dataset.

After fine-tuning, it achieves the following results on the evaluation set:
- Loss: 0.9360
- **BLEU: 55.6535**
- **ChrF++: 72.19**

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
- training_steps: 3500
- 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 | Bleu    |
|:-------------:|:-------:|:----:|:---------------:|:-------:|
| 0.2758        | 1.6667  | 250  | 0.7646          | 50.6055 |
| 0.0592        | 3.3333  | 500  | 0.7730          | 53.1258 |
| 0.0406        | 5.0     | 750  | 0.7860          | 53.3406 |
| 0.0173        | 6.6667  | 1000 | 0.8358          | 51.9789 |
| 0.0091        | 8.3333  | 1250 | 0.8909          | 54.4806 |
| 0.0071        | 10.0    | 1500 | 0.8862          | 54.2655 |
| 0.0039        | 11.6667 | 1750 | 0.9216          | 52.5119 |
| 0.0014        | 13.3333 | 2000 | 0.9281          | 54.5752 |
| 0.0013        | 15.0    | 2250 | 0.9471          | 54.5791 |
| 0.0009        | 16.6667 | 2500 | 0.9541          | 54.8725 |
| 0.0006        | 18.3333 | 2750 | 0.9614          | 53.1879 |
| 0.0006        | 20.0    | 3000 | 0.9701          | 54.6499 |
| 0.0006        | 21.6667 | 3250 | 0.9739          | 54.4341 |
| 0.0006        | 23.3333 | 3500 | 0.9747          | 54.5311 |


### Framework versions

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