File size: 4,762 Bytes
c53b4c9
 
 
0c6089d
 
 
c53b4c9
 
 
0c6089d
c53b4c9
 
0c6089d
 
c53b4c9
 
 
0c6089d
 
c53b4c9
0c6089d
 
 
c53b4c9
0c6089d
c53b4c9
 
 
 
 
0c6089d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
---
language:
- cs
license: apache-2.0
tags:
- automatic-speech-recognition
- mozilla-foundation/common_voice_8_0
- generated_from_trainer
- robust-speech-event
- xlsr-fine-tuning-week
datasets:
- common_voice
model-index:
- name: Czech comodoro Wav2Vec2 XLSR 300M CV8
  results:
  - task: 
      name: Automatic Speech Recognition 
      type: automatic-speech-recognition
    dataset:
      name: Common Voice 8
      type: mozilla-foundation/common_voice_8_0
      args: cs
    metrics:
       - name: Test WER
         type: wer
         value: 47.46
       - name: Test CER
         type: cer
         value: 10.88
---
<!-- 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-xls-r-300m-cs-cv8

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 8.0 dataset.
It achieves the following results on the evaluation set:
- WER: 0.47455377483706096
- CER: 0.10877155235645618

## Model description

Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Czech using the [Common Voice](https://huggingface.co/datasets/common_voice) dataset.
When using this model, make sure that your speech input is sampled at 16kHz.


The model can be used directly (without a language model) as follows:

```python
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

test_dataset = load_dataset("mozilla-foundation/common_voice_8_0", "cs", split="test[:2%]")

processor = Wav2Vec2Processor.from_pretrained("comodoro/wav2vec2-xls-r-300m-cs-cv8")
model = Wav2Vec2ForCTC.from_pretrained("comodoro/wav2vec2-xls-r-300m-cs-cv8")

resampler = torchaudio.transforms.Resample(48_000, 16_000)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
	speech_array, sampling_rate = torchaudio.load(batch["path"])
	batch["speech"] = resampler(speech_array).squeeze().numpy()
	return batch

test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset[:2]["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)

with torch.no_grad():
	logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits

predicted_ids = torch.argmax(logits, dim=-1)

print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset[:2]["sentence"])
```

## Evaluation

The model can be evaluated using the attached `eval.py` script.

## Training and evaluation data

The Common Voice 8.0 `train` and `validation` datasets were used for training

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 7e-05
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 20
- total_train_batch_size: 640
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 150
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch  | Step | Validation Loss | Wer    | Cer    |
|:-------------:|:------:|:----:|:---------------:|:------:|:------:|
| 7.2926        | 8.06   | 250  | 3.8497          | 1.0    | 1.0    |
| 3.417         | 16.13  | 500  | 3.2852          | 1.0    | 0.9857 |
| 2.0264        | 24.19  | 750  | 0.7099          | 0.7342 | 0.1768 |
| 0.4018        | 32.25  | 1000 | 0.6188          | 0.6415 | 0.1551 |
| 0.2444        | 40.32  | 1250 | 0.6632          | 0.6362 | 0.1600 |
| 0.1882        | 48.38  | 1500 | 0.6070          | 0.5783 | 0.1388 |
| 0.153         | 56.44  | 1750 | 0.6425          | 0.5720 | 0.1377 |
| 0.1214        | 64.51  | 2000 | 0.6363          | 0.5546 | 0.1337 |
| 0.1011        | 72.57  | 2250 | 0.6310          | 0.5222 | 0.1224 |
| 0.0879        | 80.63  | 2500 | 0.6353          | 0.5258 | 0.1253 |
| 0.0782        | 88.7   | 2750 | 0.6078          | 0.4904 | 0.1127 |
| 0.0709        | 96.76  | 3000 | 0.6465          | 0.4960 | 0.1154 |
| 0.0661        | 104.82 | 3250 | 0.6622          | 0.4945 | 0.1166 |
| 0.0616        | 112.89 | 3500 | 0.6440          | 0.4786 | 0.1104 |
| 0.0579        | 120.95 | 3750 | 0.6815          | 0.4887 | 0.1144 |
| 0.0549        | 129.03 | 4000 | 0.6603          | 0.4780 | 0.1105 |
| 0.0527        | 137.09 | 4250 | 0.6652          | 0.4749 | 0.1090 |
| 0.0506        | 145.16 | 4500 | 0.6958          | 0.4846 | 0.1133 |


### Framework versions

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