Hugging Face's logo

mozilla-ai
/
whisper-small-fr

Automatic Speech Recognition
Transformers
TensorBoard
Safetensors
French
whisper
Eval Results (legacy)
Model card Files
xet
 Metrics Community
kostissz commited on
Commit
6882cb4
·
verified ·
1 Parent(s): 55a3c03

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +20 -103
README.md CHANGED
@@ -1,118 +1,35 @@
1
  ---
2
- library_name: transformers
3
- license: apache-2.0
4
  base_model: openai/whisper-small
5
- tags:
6
- - generated_from_trainer
7
  datasets:
8
- - common_voice_17_0
9
- metrics:
10
- - wer
 
11
  model-index:
12
- - name: whisper-small-fr
13
  results:
14
  - task:
15
- name: Automatic Speech Recognition
16
  type: automatic-speech-recognition
 
17
  dataset:
18
- name: common_voice_17_0
19
- type: common_voice_17_0
20
- config: fr
21
- split: test
22
- args: fr
23
  metrics:
24
- - name: Wer
25
- type: wer
26
- value: 23.51069802258125
27
  ---
28
 
29
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
30
- should probably proofread and complete it, then remove this comment. -->
31
-
32
- # whisper-small-fr
33
-
34
- This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on the common_voice_17_0 dataset.
35
- It achieves the following results on the evaluation set:
36
- - Loss: 0.4490
37
- - Model Preparation Time: 0.0042
38
- - Wer: 23.5107
39
-
40
- ## Model description
41
-
42
- More information needed
43
-
44
- ## Intended uses & limitations
45
-
46
- More information needed
47
-
48
- ## Training and evaluation data
49
-
50
- More information needed
51
-
52
- ## Training procedure
53
-
54
- ### Training hyperparameters
55
-
56
- The following hyperparameters were used during training:
57
- - learning_rate: 1e-05
58
- - train_batch_size: 258
59
- - eval_batch_size: 64
60
- - seed: 42
61
- - optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
62
- - lr_scheduler_type: linear
63
- - lr_scheduler_warmup_steps: 50
64
- - training_steps: 2000
65
- - mixed_precision_training: Native AMP
66
-
67
- ### Training results
68
 
69
- | Training Loss | Epoch | Step | Validation Loss | Model Preparation Time | Wer |
70
- |:-------------:|:-------:|:----:|:---------------:|:----------------------:|:-------:|
71
- | 0.6515 | 0.6410 | 50 | 0.4803 | 0.0042 | 25.5359 |
72
- | 0.3297 | 1.2821 | 100 | 0.4457 | 0.0042 | 23.9349 |
73
- | 0.2733 | 1.9231 | 150 | 0.4333 | 0.0042 | 23.5294 |
74
- | 0.2003 | 2.5641 | 200 | 0.4395 | 0.0042 | 23.5419 |
75
- | 0.1763 | 3.2051 | 250 | 0.4479 | 0.0042 | 23.7748 |
76
- | 0.1354 | 3.8462 | 300 | 0.4490 | 0.0042 | 23.5107 |
77
- | 0.105 | 4.4872 | 350 | 0.4650 | 0.0042 | 23.8871 |
78
- | 0.089 | 5.1282 | 400 | 0.4833 | 0.0042 | 24.1844 |
79
- | 0.0629 | 5.7692 | 450 | 0.4929 | 0.0042 | 24.3882 |
80
- | 0.0499 | 6.4103 | 500 | 0.5136 | 0.0042 | 24.6647 |
81
- | 0.0406 | 7.0513 | 550 | 0.5226 | 0.0042 | 24.5878 |
82
- | 0.0276 | 7.6923 | 600 | 0.5368 | 0.0042 | 25.3197 |
83
- | 0.0238 | 8.3333 | 650 | 0.5504 | 0.0042 | 24.6003 |
84
- | 0.0193 | 8.9744 | 700 | 0.5593 | 0.0042 | 25.0140 |
85
- | 0.0146 | 9.6154 | 750 | 0.5718 | 0.0042 | 25.0203 |
86
- | 0.0133 | 10.2564 | 800 | 0.5816 | 0.0042 | 25.0556 |
87
- | 0.0115 | 10.8974 | 850 | 0.5867 | 0.0042 | 24.9849 |
88
- | 0.0099 | 11.5385 | 900 | 0.5946 | 0.0042 | 25.0120 |
89
- | 0.0091 | 12.1795 | 950 | 0.6006 | 0.0042 | 24.9787 |
90
- | 0.0081 | 12.8205 | 1000 | 0.6056 | 0.0042 | 25.1471 |
91
- | 0.0075 | 13.4615 | 1050 | 0.6114 | 0.0042 | 25.0972 |
92
- | 0.0072 | 14.1026 | 1100 | 0.6166 | 0.0042 | 25.0993 |
93
- | 0.0065 | 14.7436 | 1150 | 0.6198 | 0.0042 | 25.1430 |
94
- | 0.0062 | 15.3846 | 1200 | 0.6249 | 0.0042 | 25.2968 |
95
- | 0.006 | 16.0256 | 1250 | 0.6270 | 0.0042 | 25.1637 |
96
- | 0.0055 | 16.6667 | 1300 | 0.6311 | 0.0042 | 25.1741 |
97
- | 0.0053 | 17.3077 | 1350 | 0.6344 | 0.0042 | 25.2428 |
98
- | 0.0051 | 17.9487 | 1400 | 0.6371 | 0.0042 | 25.2677 |
99
- | 0.0048 | 18.5897 | 1450 | 0.6397 | 0.0042 | 25.3072 |
100
- | 0.0048 | 19.2308 | 1500 | 0.6418 | 0.0042 | 25.2532 |
101
- | 0.0046 | 19.8718 | 1550 | 0.6443 | 0.0042 | 25.3093 |
102
- | 0.0044 | 20.5128 | 1600 | 0.6460 | 0.0042 | 25.2344 |
103
- | 0.0043 | 21.1538 | 1650 | 0.6479 | 0.0042 | 25.2739 |
104
- | 0.0042 | 21.7949 | 1700 | 0.6493 | 0.0042 | 25.2802 |
105
- | 0.0042 | 22.4359 | 1750 | 0.6506 | 0.0042 | 25.3155 |
106
- | 0.0041 | 23.0769 | 1800 | 0.6519 | 0.0042 | 25.2864 |
107
- | 0.004 | 23.7179 | 1850 | 0.6528 | 0.0042 | 25.2719 |
108
- | 0.004 | 24.3590 | 1900 | 0.6531 | 0.0042 | 25.2677 |
109
- | 0.0039 | 25.0 | 1950 | 0.6538 | 0.0042 | 25.2781 |
110
- | 0.0039 | 25.6410 | 2000 | 0.6540 | 0.0042 | 25.2802 |
111
 
 
112
 
113
- ### Framework versions
 
 
114
 
115
- - Transformers 4.49.0
116
- - Pytorch 2.6.0+cu124
117
- - Datasets 3.3.2
118
- - Tokenizers 0.21.0
 
1
  ---
 
 
2
  base_model: openai/whisper-small
 
 
3
  datasets:
4
+ - mozilla-foundation/common_voice_17_0
5
+ language: fr
6
+ library_name: transformers
7
+ license: apache-2.0
8
  model-index:
9
+ - name: Finetuned openai/whisper-small on French
10
  results:
11
  - task:
 
12
  type: automatic-speech-recognition
13
+ name: Speech-to-Text
14
  dataset:
15
+ name: Common Voice (French)
16
+ type: common_voice
 
 
 
17
  metrics:
18
+ - type: wer
19
+ value: 23.511
 
20
  ---
21
 
22
+ # Finetuned openai/whisper-small on 20000 French training audio samples from mozilla-foundation/common_voice_17_0.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
 
24
+ This model was created from the Mozilla.ai Blueprint:
25
+ [speech-to-text-finetune](https://github.com/mozilla-ai/speech-to-text-finetune).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
26
 
27
+ ## Evaluation results on 5000 audio samples of French:
28
 
29
+ ### Baseline model (before finetuning) on French
30
+ - Word Error Rate: 30.304
31
+ - Loss: 1.155
32
 
33
+ ### Finetuned model (after finetuning) on French
34
+ - Word Error Rate: 23.511
35
+ - Loss: 0.449