File size: 2,984 Bytes
e7af1f8
28357e7
 
e7af1f8
 
28357e7
 
e7af1f8
b4be586
 
e7af1f8
 
 
b4be586
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7af1f8
 
 
 
 
 
 
28357e7
e7af1f8
28357e7
 
e7af1f8
b4be586
 
e7af1f8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8de96ae
e7af1f8
 
 
 
8de96ae
 
 
 
 
 
 
 
 
 
 
 
 
e7af1f8
 
 
 
 
 
 
 
b4be586
 
 
 
 
 
 
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
---
language:
- ja
license: apache-2.0
tags:
- automatic-speech-recognition
- mozilla-foundation/common_voice_8_0
- generated_from_trainer
- robust-speech-event
- ja
datasets:
- common_voice
model-index:
- name: 'XLS-R-300-m'
  results:
  - task:
      name: Automatic Speech Recognition 
      type: automatic-speech-recognition
    dataset:
      name: Common Voice 8
      type: mozilla-foundation/common_voice_8_0
      args: ja
    metrics:
       - name: Test WER
         type: wer
         value: 94.91
       - name: Test CER
         type: cer
         value: 23.32
---

<!-- 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. -->

# 

This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - JA dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5351
- Wer: 2.6188

Kanji are converted into Hiragana using the [pykakasi](https://pykakasi.readthedocs.io/en/latest/index.html) library during training and evaluation. The model can output both Hiragana and Katakana characters.

## 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: 7.5e-05
- train_batch_size: 48
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 50.0
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step  | Validation Loss | Wer    |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 4.221         | 4.5   | 1000  | 4.1195          | 2.4024 |
| 2.3597        | 9.01  | 2000  | 1.1024          | 2.7618 |
| 1.8795        | 13.51 | 3000  | 0.7498          | 2.5885 |
| 1.7143        | 18.02 | 4000  | 0.6539          | 2.5976 |
| 1.6025        | 22.52 | 5000  | 0.5989          | 2.6034 |
| 1.5403        | 27.03 | 6000  | 0.6035          | 2.6946 |
| 1.4773        | 31.53 | 7000  | 0.5647          | 2.5558 |
| 1.4228        | 36.04 | 8000  | 0.5477          | 2.5676 |
| 1.3801        | 40.54 | 9000  | 0.5413          | 2.6192 |
| 1.3558        | 45.05 | 10000 | 0.5343          | 2.6575 |
| 1.3298        | 49.55 | 11000 | 0.5349          | 2.6274 |


### Framework versions

- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu102
- Datasets 1.18.2.dev0
- Tokenizers 0.11.0

#### Evaluation Commands
1. To evaluate on `mozilla-foundation/common_voice_8_0` with split `test`

```bash
python ./eval.py --model_id AndrewMcDowell/wav2vec2-xls-r-300m-japanese --dataset mozilla-foundation/common_voice_8_0 --config ja --split test --log_outputs
```