File size: 5,701 Bytes
0756430
 
 
 
 
dde0efe
 
0756430
 
 
 
 
 
 
dde0efe
 
2b408a4
0756430
dde0efe
a018612
 
 
 
0756430
db7a68e
0756430
 
 
 
 
c1e0d6b
7a3b47d
0756430
 
 
 
dde0efe
0756430
 
 
 
 
 
 
 
 
 
dde0efe
0756430
128a1fa
 
dde0efe
 
 
 
 
 
 
c73f4cb
dde0efe
 
2b74644
dde0efe
 
cc0eebe
 
c73f4cb
 
dde0efe
 
 
 
 
 
c73f4cb
dde0efe
 
 
 
 
 
 
 
 
 
c73f4cb
dde0efe
 
 
 
 
 
 
 
 
 
c73f4cb
 
dde0efe
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c73f4cb
dde0efe
 
 
 
872496b
 
 
dde0efe
 
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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
---
language: vi
datasets:
- vivos
- common_voice
- fpt
- vlsp 100h
metrics:
- wer
pipeline_tag: automatic-speech-recognition
tags:
- audio
- speech
- Transformer
- wav2vec2
- automatic-speech-recognition
- vietnamese
license: cc-by-nc-4.0
widget:
- example_title: common_voice_vi_30519758.mp3
  src: https://huggingface.co/khanhld/wav2vec2-base-vietnamese-160h/raw/main/examples/common_voice_vi_30519758.mp3
- example_title: VIVOSDEV15_020.wav
  src: https://huggingface.co/khanhld/wav2vec2-base-vietnamese-160h/raw/main/examples/VIVOSDEV15_020.wav
model-index:
- name: Wav2vec2 Base Vietnamese 160h
  results:
  - task: 
      name: Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: common-voice-vietnamese
      type: common_voice
      args: vi
    metrics:
       - name: Test WER
         type: wer
         value: 10.78
  - task: 
      name: Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: VIVOS
      type: vivos
      args: vi
    metrics:
       - name: Test WER
         type: wer
         value: 15.05
---
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/wav2vec2-base-vietnamese-160h/speech-recognition-on-common-voice-vi)](https://paperswithcode.com/sota/speech-recognition-on-common-voice-vi?p=wav2vec2-base-vietnamese-160h)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/wav2vec2-base-vietnamese-160h/speech-recognition-on-vivos)](https://paperswithcode.com/sota/speech-recognition-on-vivos?p=wav2vec2-base-vietnamese-160h)
# Vietnamese Speech Recognition using Wav2vec 2.0
### Table of contents
1. [Model Description](#description)
2. [Benchmark Result](#benchmark)
3. [Example Usage](#example)
4. [Evaluation](#evaluation)
5. [Contact](#contact)

<a name = "description" ></a>
### Model Description
Fine-tuned the Wav2vec2-based model on about 160 hours of Vietnamese speech dataset from different resources including [VIOS](https://huggingface.co/datasets/vivos), [COMMON VOICE](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0), [FOSD](https://data.mendeley.com/datasets/k9sxg2twv4/4) and [VLSP 100h](https://drive.google.com/file/d/1vUSxdORDxk-ePUt-bUVDahpoXiqKchMx/view). We have not yet incorporated the Language Model into our ASR system but still gained a promising result.
<br>
We also provide code for Pre-training and Fine-tuning the Wav2vec2 model (not available for now but will release soon). If you wish to train on your dataset, check it out here:
- [Pretrain](https://github.com/khanld/ASR-Wav2vec-Pretrain)
- [Finetune](https://github.com/khanld/ASR-Wa2vec-Finetune)
</br>

<a name = "benchmark" ></a>
### Benchmark WER Result
| | [VIVOS](https://huggingface.co/datasets/vivos) | [COMMON VOICE 8.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0) | 
|---|---|---|
|without LM| 15.05 | 10.78 |
|with LM| in progress | in progress |

<a name = "example" ></a>
### Example Usage
```python
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
import librosa
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

processor = Wav2Vec2Processor.from_pretrained("khanhld/wav2vec2-base-vietnamese-160h")
model = Wav2Vec2ForCTC.from_pretrained("khanhld/wav2vec2-base-vietnamese-160h")
model.to(device)

def transcribe(wav):
  input_values = processor(wav, sampling_rate=16000, return_tensors="pt").input_values
  logits = model(input_values.to(device)).logits
  pred_ids = torch.argmax(logits, dim=-1)
  pred_transcript = processor.batch_decode(pred_ids)[0]
  return pred_transcript


wav, _ = librosa.load('path/to/your/audio/file', sr = 16000)
print(f"transcript: {transcribe(wav)}")
```

<a name = "evaluation"></a>
### Evaluation
```python
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
from datasets import load_dataset
import torch
import re
from datasets import load_dataset, load_metric, Audio

wer = load_metric("wer")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# load processor and model
processor = Wav2Vec2Processor.from_pretrained("khanhld/wav2vec2-base-vietnamese-160h")
model = Wav2Vec2ForCTC.from_pretrained("khanhld/wav2vec2-base-vietnamese-160h")
model.to(device)
model.eval()

# Load dataset
test_dataset = load_dataset("mozilla-foundation/common_voice_8_0", "vi", split="test")
test_dataset = test_dataset.cast_column("audio", Audio(sampling_rate=16000))
chars_to_ignore = r'[,?.!\-;:"“%\'�]' # ignore special characters

# preprocess data
def preprocess(batch):
  audio = batch["audio"]
  batch["input_values"] = audio["array"]
  batch["transcript"] = re.sub(chars_to_ignore, '', batch["sentence"]).lower()
  return batch

# run inference
def inference(batch):
  input_values = processor(batch["input_values"], 
                            sampling_rate=16000, 
                            return_tensors="pt").input_values
  logits = model(input_values.to(device)).logits
  pred_ids = torch.argmax(logits, dim=-1)
  batch["pred_transcript"] = processor.batch_decode(pred_ids) 
  return batch
  
test_dataset = test_dataset.map(preprocess)
result = test_dataset.map(inference, batched=True, batch_size=1)
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_transcript"], references=result["transcript"])))
```
**Test Result**: 10.78%

<a name = "contact"></a>
### Contact
- [email protected]
- [![GitHub](https://img.shields.io/badge/github-%23121011.svg?style=for-the-badge&logo=github&logoColor=white)](https://github.com/)
- [![LinkedIn](https://img.shields.io/badge/linkedin-%230077B5.svg?style=for-the-badge&logo=linkedin&logoColor=white)](https://www.linkedin.com/in/khanhld257/)