Hugging Face's logo

comodoro
/
wav2vec2-xls-r-300m-west-slavic-cv8

Automatic Speech Recognition
Transformers
PyTorch
5 languages
wav2vec2
Generated from Trainer
hf-asr-leaderboard
mozilla-foundation/common_voice_8_0
robust-speech-event
xlsr-fine-tuning-week
Eval Results
Inference Endpoints
Model card Files Community
1
comodoro commited on
Commit
fb2ee5e
•
1 Parent(s): 0f3175a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +88 -67
README.md CHANGED
@@ -1,37 +1,104 @@
1
  ---
 
 
 
 
 
 
2
  license: apache-2.0
3
  tags:
 
 
4
  - generated_from_trainer
 
 
5
  model-index:
6
  - name: wav2vec2-xls-r-300m-west-slavic-cv8
7
- results: []
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8
  ---
9
 
10
- <!-- This model card has been generated automatically according to the information the Trainer had access to. You
11
- should probably proofread and complete it, then remove this comment. -->
12
-
13
  # wav2vec2-xls-r-300m-west-slavic-cv8
14
 
15
- This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the None dataset.
16
- It achieves the following results on the evaluation set:
17
- - Loss: 2.3462
18
- - Wer: 0.8556
19
- - Cer: 0.2799
20
-
21
- ## Model description
22
-
23
- More information needed
24
-
25
- ## Intended uses & limitations
26
-
27
- More information needed
28
 
29
- ## Training and evaluation data
30
 
31
- More information needed
32
-
33
- ## Training procedure
34
 
 
 
 
 
35
  ### Training hyperparameters
36
 
37
  The following hyperparameters were used during training:
@@ -45,52 +112,6 @@ The following hyperparameters were used during training:
45
  - num_epochs: 50
46
  - mixed_precision_training: Native AMP
47
 
48
- ### Training results
49
-
50
- | Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
51
- |:-------------:|:-----:|:-----:|:---------------:|:------:|:------:|
52
- | 6.548 | 1.23 | 400 | 3.4763 | 1.0 | 1.0 |
53
- | 3.42 | 2.45 | 800 | 3.3156 | 1.0 | 1.0 |
54
- | 3.291 | 3.68 | 1200 | 3.2396 | 1.0 | 1.0 |
55
- | 2.6515 | 4.91 | 1600 | 2.0422 | 0.9997 | 0.5835 |
56
- | 1.7019 | 6.13 | 2000 | 1.6337 | 0.9893 | 0.4797 |
57
- | 1.3604 | 7.36 | 2400 | 1.5221 | 0.9875 | 0.4463 |
58
- | 1.1965 | 8.59 | 2800 | 1.5284 | 0.9766 | 0.4247 |
59
- | 1.069 | 9.82 | 3200 | 1.5228 | 0.9672 | 0.4124 |
60
- | 0.9536 | 11.04 | 3600 | 1.4059 | 0.9600 | 0.3868 |
61
- | 0.8487 | 12.27 | 4000 | 1.4083 | 0.9501 | 0.3739 |
62
- | 0.7655 | 13.5 | 4400 | 1.4079 | 0.9369 | 0.3612 |
63
- | 0.6956 | 14.72 | 4800 | 1.4170 | 0.9411 | 0.3459 |
64
- | 0.6287 | 15.95 | 5200 | 1.4000 | 0.9235 | 0.3384 |
65
- | 0.561 | 17.18 | 5600 | 1.4735 | 0.9023 | 0.3295 |
66
- | 0.5155 | 18.4 | 6000 | 1.5386 | 0.9202 | 0.3223 |
67
- | 0.4864 | 19.63 | 6400 | 1.6186 | 0.9073 | 0.3259 |
68
- | 0.4261 | 20.86 | 6800 | 1.6417 | 0.9217 | 0.3130 |
69
- | 0.4051 | 22.09 | 7200 | 1.6295 | 0.8954 | 0.3026 |
70
- | 0.3779 | 23.31 | 7600 | 1.8218 | 0.8979 | 0.3153 |
71
- | 0.35 | 24.54 | 8000 | 1.7790 | 0.8921 | 0.3036 |
72
- | 0.3343 | 25.77 | 8400 | 1.8588 | 0.9114 | 0.3072 |
73
- | 0.3137 | 26.99 | 8800 | 1.8096 | 0.8756 | 0.2935 |
74
- | 0.299 | 28.22 | 9200 | 1.9721 | 0.8863 | 0.3023 |
75
- | 0.2894 | 29.45 | 9600 | 1.9907 | 0.8872 | 0.2958 |
76
- | 0.2784 | 30.67 | 10000 | 1.9494 | 0.9090 | 0.2945 |
77
- | 0.2662 | 31.9 | 10400 | 1.9952 | 0.8978 | 0.2935 |
78
- | 0.2614 | 33.13 | 10800 | 2.0600 | 0.8949 | 0.2979 |
79
- | 0.2401 | 34.36 | 11200 | 2.1180 | 0.8914 | 0.2950 |
80
- | 0.2392 | 35.58 | 11600 | 2.1197 | 0.8713 | 0.2895 |
81
- | 0.23 | 36.81 | 12000 | 2.1680 | 0.8713 | 0.2941 |
82
- | 0.2246 | 38.04 | 12400 | 2.1526 | 0.8741 | 0.2879 |
83
- | 0.2152 | 39.26 | 12800 | 2.2631 | 0.8790 | 0.2889 |
84
- | 0.212 | 40.49 | 13200 | 2.2724 | 0.8661 | 0.2843 |
85
- | 0.2044 | 41.72 | 13600 | 2.2438 | 0.8691 | 0.2878 |
86
- | 0.2029 | 42.94 | 14000 | 2.2519 | 0.8577 | 0.2833 |
87
- | 0.1972 | 44.17 | 14400 | 2.2697 | 0.8604 | 0.2813 |
88
- | 0.1884 | 45.4 | 14800 | 2.3294 | 0.8662 | 0.2847 |
89
- | 0.1877 | 46.63 | 15200 | 2.3077 | 0.8561 | 0.2793 |
90
- | 0.1871 | 47.85 | 15600 | 2.3518 | 0.8563 | 0.2801 |
91
- | 0.1838 | 49.08 | 16000 | 2.3462 | 0.8556 | 0.2799 |
92
-
93
-
94
  ### Framework versions
95
 
96
  - Transformers 4.16.0.dev0
 
1
  ---
2
+ language:
3
+ - cs
4
+ - hsb
5
+ - pl
6
+ - sk
7
+ - sl
8
  license: apache-2.0
9
  tags:
10
+ - automatic-speech-recognition
11
+ - mozilla-foundation/common_voice_8_0
12
  - generated_from_trainer
13
+ - robust-speech-event
14
+ - xlsr-fine-tuning-week
15
  model-index:
16
  - name: wav2vec2-xls-r-300m-west-slavic-cv8
17
+ results:
18
+ - task:
19
+ name: Automatic Speech Recognition
20
+ type: automatic-speech-recognition
21
+ dataset:
22
+ name: Common Voice 8
23
+ type: mozilla-foundation/common_voice_8_0
24
+ args: cs
25
+ metrics:
26
+ - name: Test WER
27
+ type: wer
28
+ value: 53.5
29
+ - name: Test CER
30
+ type: cer
31
+ value: 14.7
32
+ - task:
33
+ name: Automatic Speech Recognition
34
+ type: automatic-speech-recognition
35
+ dataset:
36
+ name: Common Voice 8
37
+ type: mozilla-foundation/common_voice_8_0
38
+ args: hsb
39
+ metrics:
40
+ - name: Test WER
41
+ type: wer
42
+ value: 81.7
43
+ - name: Test CER
44
+ type: cer
45
+ value: 21.2
46
+ - task:
47
+ name: Automatic Speech Recognition
48
+ type: automatic-speech-recognition
49
+ dataset:
50
+ name: Common Voice 8
51
+ type: mozilla-foundation/common_voice_8_0
52
+ args: pl
53
+ metrics:
54
+ - name: Test WER
55
+ type: wer
56
+ value: 60.2
57
+ - name: Test CER
58
+ type: cer
59
+ value: 15.6
60
+ - task:
61
+ name: Automatic Speech Recognition
62
+ type: automatic-speech-recognition
63
+ dataset:
64
+ name: Common Voice 8
65
+ type: mozilla-foundation/common_voice_8_0
66
+ args: sk
67
+ metrics:
68
+ - name: Test WER
69
+ type: wer
70
+ value: 69.6
71
+ - name: Test CER
72
+ type: cer
73
+ value: 20.7
74
+ - task:
75
+ name: Automatic Speech Recognition
76
+ type: automatic-speech-recognition
77
+ dataset:
78
+ name: Common Voice 8
79
+ type: mozilla-foundation/common_voice_8_0
80
+ args: sl
81
+ metrics:
82
+ - name: Test WER
83
+ type: wer
84
+ value: 73.2
85
+ - name: Test CER
86
+ type: cer
87
+ value: 23.2
88
  ---
89
 
 
 
 
90
  # wav2vec2-xls-r-300m-west-slavic-cv8
91
 
92
+ 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 dataset of five similar languages with similar scripts: Czech, Slovak, Polish, Slovenian and Upper Sorbian. Training and validation sets were concatenated and shuffled.
 
 
 
 
 
 
 
 
 
 
 
 
93
 
94
+ Evaluation set used for training was concatenated from the respective test sets and shuffled while limiting each language to at most 2000 samples. During training, cca WER 70 was achieved on this set.
95
 
96
+ ### Evaluation script
 
 
97
 
98
+ ```
99
+ python eval.py --model_id comodoro/wav2vec2-xls-r-300m-west-slavic-cv8 --dataset mozilla-foundation/common_voice_8_0 --split test --config {lang}
100
+ ```
101
+
102
  ### Training hyperparameters
103
 
104
  The following hyperparameters were used during training:
 
112
  - num_epochs: 50
113
  - mixed_precision_training: Native AMP
114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
115
  ### Framework versions
116
 
117
  - Transformers 4.16.0.dev0