File size: 3,852 Bytes
193880e
 
20d85b6
 
193880e
 
20d85b6
 
193880e
 
20d85b6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
193880e
 
 
 
 
 
 
20d85b6
193880e
0a873ad
 
 
 
193880e
0a873ad
 
 
 
193880e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0a873ad
 
 
 
 
 
 
193880e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
---
license: apache-2.0
language:
- vi
tags:
- automatic-speech-recognition
- robust-speech-event
- common-voice
model-index:
- name: xls-asr-vi-40h-1B
  results:
  - task: 
      name: Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: Common Voice 7.0 vi
      type: common_voice
      args: vi
    metrics:
       - name: Test WER
         type: wer
         value: 34.21
       - name: Test CER
         type: cer
         value: 19.94
---

<!-- 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. -->

# xls-asr-vi-40h-1B

This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the common voice 7.0 vi & private dataset.

### Benchmark WER result:
| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 7.0 VI](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) |
|---|---|---|
|without LM| 25.92 | 34.21 |

### Benchmark CER result:
| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 7.0 VI](https://huggingface.co/datasets/mozilla-foundation/common_voice_7_0) |
|---|---|---|
|without LM| 9.24 | 19.94 |

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1500
- num_epochs: 50.0
- mixed_precision_training: Native AMP
- attention_dropout: 0.2
- activation_dropout: 0.1
- warmup_steps: 1500
- mask_time_prob: .15
- mask_time_length: 10
- mask_feature_prob: 0.25
- mask_feature_length: 64

### Training results

| Training Loss | Epoch | Step  | Validation Loss | Wer    |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 4.6222        | 1.85  | 1500  | 5.9479          | 0.5474 |
| 1.1362        | 3.7   | 3000  | 7.9799          | 0.5094 |
| 0.7814        | 5.56  | 4500  | 5.0330          | 0.4724 |
| 0.6281        | 7.41  | 6000  | 2.3484          | 0.5020 |
| 0.5472        | 9.26  | 7500  | 2.2495          | 0.4793 |
| 0.4827        | 11.11 | 9000  | 1.1530          | 0.4768 |
| 0.4327        | 12.96 | 10500 | 1.6160          | 0.4646 |
| 0.3989        | 14.81 | 12000 | 3.2633          | 0.4703 |
| 0.3522        | 16.67 | 13500 | 2.2337          | 0.4708 |
| 0.3201        | 18.52 | 15000 | 3.6879          | 0.4565 |
| 0.2899        | 20.37 | 16500 | 5.4389          | 0.4599 |
| 0.2776        | 22.22 | 18000 | 3.5284          | 0.4537 |
| 0.2574        | 24.07 | 19500 | 2.1759          | 0.4649 |
| 0.2378        | 25.93 | 21000 | 3.3901          | 0.4448 |
| 0.217         | 27.78 | 22500 | 1.1632          | 0.4565 |
| 0.2115        | 29.63 | 24000 | 1.7441          | 0.4232 |
| 0.1959        | 31.48 | 25500 | 3.4992          | 0.4304 |
| 0.187         | 33.33 | 27000 | 3.6163          | 0.4369 |
| 0.1748        | 35.19 | 28500 | 3.6038          | 0.4467 |
| 0.17          | 37.04 | 30000 | 2.9708          | 0.4362 |
| 0.159         | 38.89 | 31500 | 3.2045          | 0.4279 |
| 0.153         | 40.74 | 33000 | 3.2427          | 0.4287 |
| 0.1463        | 42.59 | 34500 | 3.5439          | 0.4270 |
| 0.139         | 44.44 | 36000 | 3.9381          | 0.4150 |
| 0.1352        | 46.3  | 37500 | 4.1744          | 0.4092 |
| 0.1369        | 48.15 | 39000 | 4.2279          | 0.4154 |
| 0.1273        | 50.0  | 40500 | 4.1691          | 0.4133 |


### Framework versions

- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.11.0