Cahya Wirawan
updated the link to synthetic file c0a9750
1
---
2
language: id
3
datasets:
4
- common_voice 
5
metrics:
6
- wer
7
tags:
8
- audio
9
- automatic-speech-recognition
10
- speech
11
- xlsr-fine-tuning-week
12
license: apache-2.0
13
model-index:
14
- name: XLSR Wav2Vec2 Indonesian by Indonesian NLP
15
  results:
16
  - task: 
17
      name: Speech Recognition
18
      type: automatic-speech-recognition
19
    dataset:
20
      name: Common Voice id
21
      type: common_voice
22
      args: id
23
    metrics:
24
       - name: Test WER
25
         type: wer
26
         value: 14.29
27
---
28
29
# Wav2Vec2-Large-XLSR-Indonesian
30
31
This is the model for Wav2Vec2-Large-XLSR-Indonesian, a fine-tuned 
32
[facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53)
33
model on the [Indonesian Common Voice dataset](https://huggingface.co/datasets/common_voice).
34
When using this model, make sure that your speech input is sampled at 16kHz.
35
36
## Usage
37
The model can be used directly (without a language model) as follows:
38
```python
39
import torch
40
import torchaudio
41
from datasets import load_dataset
42
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
43
44
test_dataset = load_dataset("common_voice", "id", split="test[:2%]")
45
46
processor = Wav2Vec2Processor.from_pretrained("indonesian-nlp/wav2vec2-large-xlsr-indonesian")
47
model = Wav2Vec2ForCTC.from_pretrained("indonesian-nlp/wav2vec2-large-xlsr-indonesian")
48
49
50
# Preprocessing the datasets.
51
# We need to read the aduio files as arrays
52
def speech_file_to_array_fn(batch):
53
  speech_array, sampling_rate = torchaudio.load(batch["path"])
54
  resampler = torchaudio.transforms.Resample(sampling_rate, 16_000)
55
  batch["speech"] = resampler(speech_array).squeeze().numpy()
56
  return batch
57
58
test_dataset = test_dataset.map(speech_file_to_array_fn)
59
inputs = processor(test_dataset[:2]["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
60
61
with torch.no_grad():
62
  logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
63
64
predicted_ids = torch.argmax(logits, dim=-1)
65
66
print("Prediction:", processor.batch_decode(predicted_ids))
67
print("Reference:", test_dataset[:2]["sentence"])
68
```
69
70
71
## Evaluation
72
73
The model can be evaluated as follows on the Indonesian test data of Common Voice.
74
75
```python
76
import torch
77
import torchaudio
78
from datasets import load_dataset, load_metric
79
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
80
import re
81
82
test_dataset = load_dataset("common_voice", "id", split="test")
83
wer = load_metric("wer")
84
85
processor = Wav2Vec2Processor.from_pretrained("indonesian-nlp/wav2vec2-large-xlsr-indonesian")
86
model = Wav2Vec2ForCTC.from_pretrained("indonesian-nlp/wav2vec2-large-xlsr-indonesian") 
87
model.to("cuda")
88
89
chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\'\”\�]'
90
91
92
# Preprocessing the datasets.
93
# We need to read the aduio files as arrays
94
def speech_file_to_array_fn(batch):
95
  batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
96
  speech_array, sampling_rate = torchaudio.load(batch["path"])
97
  resampler = torchaudio.transforms.Resample(sampling_rate, 16_000)
98
  batch["speech"] = resampler(speech_array).squeeze().numpy()
99
  return batch
100
101
test_dataset = test_dataset.map(speech_file_to_array_fn)
102
103
# Preprocessing the datasets.
104
# We need to read the aduio files as arrays
105
def evaluate(batch):
106
  inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
107
108
  with torch.no_grad():
109
    logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
110
111
  pred_ids = torch.argmax(logits, dim=-1)
112
  batch["pred_strings"] = processor.batch_decode(pred_ids)
113
  return batch
114
115
result = test_dataset.map(evaluate, batched=True, batch_size=8)
116
117
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
118
```
119
120
**Test Result**: 14.29 %
121
122
## Training
123
124
The Common Voice `train`, `validation`, and [synthetic voice datasets](https://cloud.uncool.ai/index.php/s/Kg4C6f5NJGN9ZdR) were used for training.
125
126
The script used for training can be found [here](https://github.com/indonesian-nlp/wav2vec2-indonesian) 
127