File size: 2,560 Bytes
4002d6e
 
 
 
 
 
0d7452e
4002d6e
 
 
 
 
 
 
 
 
 
5f410f2
 
 
4002d6e
 
 
 
 
 
562300a
 
4002d6e
 
 
 
 
 
 
dab015f
 
4002d6e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
79
80
81
82
83
---
license: apache-2.0
tags:
- generated_from_trainer
metrics:
- wer
base_model: openai/whisper-medium
model-index:
- name: whisper-atcosim3
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# whisper-atcosim3

This model is a fine-tuned version of [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) 
on the [atco2_atcosim](https://huggingface.co/datasets/luigisaetta/atco2_atcosim) 
dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0476
- Wer: 0.0198

## Model description

This model is a special ASR model, derived doing fine-tuning of OpenAI Whisper model on ATC conversations.
The base model is: [OpenAI Whisper Medium](https://huggingface.co/openai/whisper-medium)

## Intended uses & limitations

More information needed

## Training and evaluation data

The model has been trained on the [atco2_atcosim](https://huggingface.co/datasets/luigisaetta/atco2_atcosim)
dataset.

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- gradient_accumulation_steps: 8
- total_train_batch_size: 32
- total_eval_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 150
- training_steps: 600
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Wer    |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.8218        | 0.2   | 50   | 0.3785          | 0.1451 |
| 0.1429        | 0.39  | 100  | 0.1213          | 0.0714 |
| 0.1155        | 0.59  | 150  | 0.0807          | 0.0517 |
| 0.0764        | 0.79  | 200  | 0.0652          | 0.0272 |
| 0.0724        | 0.98  | 250  | 0.0607          | 0.0393 |
| 0.0357        | 1.18  | 300  | 0.0569          | 0.0242 |
| 0.03          | 1.38  | 350  | 0.0553          | 0.0243 |
| 0.0325        | 1.57  | 400  | 0.0556          | 0.0228 |
| 0.03          | 1.77  | 450  | 0.0501          | 0.0242 |
| 0.0232        | 1.97  | 500  | 0.0485          | 0.0205 |
| 0.0143        | 2.16  | 550  | 0.0480          | 0.0194 |
| 0.0105        | 2.36  | 600  | 0.0476          | 0.0198 |


### Framework versions

- Transformers 4.29.0
- Pytorch 2.0.0+cu117
- Datasets 2.12.0
- Tokenizers 0.11.0