File size: 6,075 Bytes
b9072de
 
02a26b0
 
 
b9072de
 
 
 
02a26b0
 
b9072de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
02a26b0
 
 
 
 
 
 
 
 
 
 
 
 
 
b9072de
 
 
 
2b478a5
 
 
 
 
 
 
b9072de
02a26b0
b9072de
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
---
license: apache-2.0
language: fr
library_name: transformers
thumbnail: null
tags:
- automatic-speech-recognition
- hf-asr-leaderboard
- robust-speech-event
- CTC
- Wav2vec2
datasets:
- common_voice
- mozilla-foundation/common_voice_11_0
- facebook/multilingual_librispeech
- facebook/voxpopuli
- gigant/african_accented_french
metrics:
- wer
model-index:
- name: Fine-tuned wav2vec2-FR-7K-large model for ASR in French
  results:
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: Common Voice 11
      type: mozilla-foundation/common_voice_11_0
      args: fr
    metrics:
    - name: Test WER
      type: wer
      value: 11.44
    - name: Test WER (+LM)
      type: wer
      value: 9.66
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: Multilingual LibriSpeech (MLS)
      type: facebook/multilingual_librispeech
      args: french
    metrics:
    - name: Test WER
      type: wer
      value: 5.93
    - name: Test WER (+LM)
      type: wer
      value: 5.13
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: VoxPopuli
      type: facebook/voxpopuli
      args: fr
    metrics:
    - name: Test WER
      type: wer
      value: 9.33
    - name: Test WER (+LM)
      type: wer
      value: 8.51
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: African Accented French
      type: gigant/african_accented_french
      args: fr
    metrics:
    - name: Test WER
      type: wer
      value: 16.22
    - name: Test WER (+LM)
      type: wer
      value: 15.39
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: Robust Speech Event - Dev Data
      type: speech-recognition-community-v2/dev_data
      args: fr
    metrics:
    - name: Test WER
      type: wer
      value: 16.56
    - name: Test WER (+LM)
      type: wer
      value: 12.96
  - task:
      name: Automatic Speech Recognition
      type: automatic-speech-recognition
    dataset:
      name: Fleurs
      type: google/fleurs
      args: fr_fr
    metrics:
    - name: Test WER
      type: wer
      value: 10.10
    - name: Test WER (+LM)
      type: wer
      value: 8.84
---

# Fine-tuned wav2vec2-FR-7K-large model for ASR in French

<style>
img {
 display: inline;
}
</style>

[![Model architecture](https://img.shields.io/badge/Model_Architecture-Wav2Vec2--CTC-lightgrey)](#model-architecture) [![Model size](https://img.shields.io/badge/Params-315M-lightgrey)](#model-architecture) [![Language](https://img.shields.io/badge/Language-French-lightgrey)](#datasets)

This model is a fine-tuned version of [LeBenchmark/wav2vec2-FR-7K-large](https://huggingface.co/LeBenchmark/wav2vec2-FR-7K-large), trained on a composite dataset comprising of over 2200 hours of French speech audio, using the train and validation splits of [Common Voice 11.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0), [Multilingual LibriSpeech](https://huggingface.co/datasets/facebook/multilingual_librispeech), [Voxpopuli](https://github.com/facebookresearch/voxpopuli), [Multilingual TEDx](http://www.openslr.org/100), [MediaSpeech](https://www.openslr.org/108), and [African Accented French](https://huggingface.co/datasets/gigant/african_accented_french). When using the model make sure that your speech input is also sampled at 16Khz.

## Usage

1. To use on a local audio file with the language model

```python
import torch
import torchaudio

from transformers import AutoModelForCTC, Wav2Vec2ProcessorWithLM

model = AutoModelForCTC.from_pretrained("bhuang/asr-wav2vec2-french").cuda()
processor_with_lm = Wav2Vec2ProcessorWithLM.from_pretrained("bhuang/asr-wav2vec2-french")

wav_path = "example.wav"  # path to your audio file
waveform, sample_rate = torchaudio.load(wav_path)
waveform = waveform.squeeze(axis=0)  # mono

# resample
if sample_rate != 16_000:
    resampler = torchaudio.transforms.Resample(sample_rate, 16_000)
    waveform = resampler(waveform)

# normalize
input_dict = processor_with_lm(waveform, sampling_rate=16_000, return_tensors="pt")

with torch.inference_mode():
    logits = model(input_dict.input_values.to("cuda")).logits

predicted_sentence = processor_with_lm.batch_decode(logits.cpu().numpy()).text[0]
```

2. To use on a local audio file without the language model

```python
import torch
import torchaudio

from transformers import AutoModelForCTC, Wav2Vec2Processor

model = AutoModelForCTC.from_pretrained("bhuang/asr-wav2vec2-french").cuda()
processor = Wav2Vec2Processor.from_pretrained("bhuang/asr-wav2vec2-french")

wav_path = "example.wav"  # path to your audio file
waveform, sample_rate = torchaudio.load(wav_path)
waveform = waveform.squeeze(axis=0)  # mono

# resample
if sample_rate != 16_000:
    resampler = torchaudio.transforms.Resample(sample_rate, 16_000)
    waveform = resampler(waveform)

# normalize
input_dict = processor(waveform, sampling_rate=16_000, return_tensors="pt")

with torch.inference_mode():
    logits = model(input_dict.input_values.to("cuda")).logits

# decode
predicted_ids = torch.argmax(logits, dim=-1)
predicted_sentence = processor.batch_decode(predicted_ids)[0]
```

## Evaluation

1. To evaluate on `mozilla-foundation/common_voice_11_0`

```bash
python eval.py \
  --model_id "bhuang/asr-wav2vec2-french" \
  --dataset "mozilla-foundation/common_voice_11_0" \
  --config "fr" \
  --split "test" \
  --log_outputs \
  --outdir "outputs/results_mozilla-foundatio_common_voice_11_0_with_lm"
```

2. To evaluate on `speech-recognition-community-v2/dev_data`

```bash
python eval.py \
  --model_id "bhuang/asr-wav2vec2-french" \
  --dataset "speech-recognition-community-v2/dev_data" \
  --config "fr" \
  --split "validation" \
  --chunk_length_s 30.0 \
  --stride_length_s 5.0 \
  --log_outputs \
  --outdir "outputs/results_speech-recognition-community-v2_dev_data_with_lm"
```