File size: 2,485 Bytes
eea89e7
c60ce76
 
 
a2f088d
c60ce76
 
 
 
 
 
 
 
7b86932
a2f088d
c60ce76
 
 
 
 
 
 
a2f088d
7b86932
a2f088d
eea89e7
 
c60ce76
 
eea89e7
c60ce76
eea89e7
c60ce76
 
7b86932
 
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
eea89e7
c60ce76
 
 
 
 
 
 
 
 
eea89e7
c60ce76
eea89e7
c60ce76
 
7b86932
 
 
 
 
 
 
 
 
 
 
 
eea89e7
 
c60ce76
eea89e7
c60ce76
 
 
 
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
84
85
86
87
88
---
license: apache-2.0
tags:
- generated_from_trainer
base_model: facebook/wav2vec2-xls-r-300m
datasets:
- common_voice_13_0
metrics:
- wer
model-index:
- name: wav2vec2-xlsr-53-CV-demo-google-colab-Ezra_William_Prod17
  results:
  - task:
      type: automatic-speech-recognition
      name: Automatic Speech Recognition
    dataset:
      name: common_voice_13_0
      type: common_voice_13_0
      config: id
      split: test
      args: id
    metrics:
    - type: wer
      value: 0.3124539085545723
      name: Wer
---

<!-- 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. -->

# wav2vec2-xlsr-53-CV-demo-google-colab-Ezra_William_Prod17

This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the common_voice_13_0 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3690
- Wer: 0.3125

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 12
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Wer    |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 2.9121        | 1.0   | 278  | 2.8104          | 1.0    |
| 0.7904        | 2.0   | 556  | 0.5906          | 0.5911 |
| 0.4628        | 3.0   | 834  | 0.4488          | 0.5157 |
| 0.3279        | 4.0   | 1112 | 0.4175          | 0.4261 |
| 0.253         | 5.0   | 1390 | 0.3738          | 0.3929 |
| 0.1969        | 6.0   | 1668 | 0.3810          | 0.3847 |
| 0.1657        | 7.0   | 1946 | 0.3701          | 0.3587 |
| 0.1444        | 8.0   | 2224 | 0.3681          | 0.3457 |
| 0.1305        | 9.0   | 2502 | 0.3632          | 0.3229 |
| 0.1179        | 10.0  | 2780 | 0.3620          | 0.3225 |
| 0.1037        | 11.0  | 3058 | 0.3697          | 0.3136 |
| 0.0988        | 12.0  | 3336 | 0.3690          | 0.3125 |


### Framework versions

- Transformers 4.40.0
- Pytorch 2.2.2+cu121
- Datasets 2.19.0
- Tokenizers 0.19.1