File size: 3,479 Bytes
262fd2e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7c5a79
 
 
262fd2e
 
 
 
 
 
 
dbf96ee
 
7e97291
 
dbf96ee
 
262fd2e
 
 
 
 
 
 
73a10d3
262fd2e
 
 
73a10d3
 
 
 
 
 
 
262fd2e
 
 
73a10d3
262fd2e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7c5a79
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
---
license: apache-2.0
base_model: facebook/wav2vec2-large-xlsr-53
tags:
- generated_from_trainer
datasets:
- common_voice_13_0
metrics:
- wer
model-index:
- name: wav2vec2-large-xlsr-common_voice_13_0-id
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: common_voice_13_0
      type: common_voice_13_0
      config: id
      split: test
      args: id
    metrics:
    - name: Wer
      type: wer
      value: 0.4316463864306785
language:
- id
library_name: transformers
---

<!-- 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-large-xlsr-common_voice_13_0-id

> **Note:** do not recommended to try the model through this model card
>
> Alternatively, try it through the available space [click here](https://huggingface.co/spaces/arifagustyawan/wav2vec2-large-xlsr-53-id)
> Then you can addapt the inference method available in the gradio app script. Or you can checkout at my github repository [click here](https://github.com/agustyawan-arif/wav2vec2-large-xlsr-53-id)


This model is a fine-tuned version of [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on the common_voice_13_0 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4115
- Wer: 0.4316

## Model description

The model is based on the facebook/wav2vec2-large-xlsr-53 architecture and fine-tuned for Automatic Speech Recognition on the common_voice_13_0 dataset in Indonesian (id). It is designed to transcribe spoken language into written text.

## Intended uses & limitations

**Intended Uses:**
- Automatic Speech Recognition for Indonesian speech data.
- Transcription of spoken content in common_voice_13_0 dataset.

**Limitations:**
- The model's performance may vary on speech data outside the common_voice_13_0 dataset.
- It may not perform well on languages other than Indonesian.

## Training and evaluation data

The model was trained on the common_voice_13_0 dataset, specifically using the Indonesian (id) split for evaluation.

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- 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: 500
- num_epochs: 30
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Wer    |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 5.0656        | 2.88  | 400  | 2.7637          | 1.0    |
| 1.1404        | 5.76  | 800  | 0.4483          | 0.6088 |
| 0.3698        | 8.63  | 1200 | 0.4029          | 0.5278 |
| 0.2695        | 11.51 | 1600 | 0.3976          | 0.5036 |
| 0.2074        | 14.39 | 2000 | 0.3988          | 0.4793 |
| 0.1796        | 17.27 | 2400 | 0.3952          | 0.4590 |
| 0.1523        | 20.14 | 2800 | 0.3986          | 0.4463 |
| 0.1352        | 23.02 | 3200 | 0.4143          | 0.4374 |
| 0.121         | 25.9  | 3600 | 0.4022          | 0.4337 |
| 0.1085        | 28.78 | 4000 | 0.4115          | 0.4316 |


### Framework versions

- Transformers 4.35.2
- Pytorch 2.1.0+cu118
- Datasets 2.15.0
- Tokenizers 0.15.0