Akashpb13 commited on
Commit
a6f874f
1 Parent(s): 5ad8759

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +137 -0
README.md ADDED
@@ -0,0 +1,137 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - ckb
4
+ license: apache-2.0
5
+ tags:
6
+ - automatic-speech-recognition
7
+ - mozilla-foundation/common_voice_8_0
8
+ - generated_from_trainer
9
+ - ckb
10
+ - robust-speech-event
11
+ - model_for_talk
12
+ datasets:
13
+ - mozilla-foundation/common_voice_8_0
14
+
15
+ model-index:
16
+ - name: Akashpb13/Central_kurdish_xlsr
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: ckb
25
+ metrics:
26
+ - name: Test WER
27
+ type: wer
28
+ value: 0.36754389884276845
29
+ - name: Test CER
30
+ type: cer
31
+ value: 0.07827896768334217
32
+ - task:
33
+ name: Automatic Speech Recognition
34
+ type: automatic-speech-recognition
35
+ dataset:
36
+ name: Robust Speech Event - Dev Data
37
+ type: speech-recognition-community-v2/dev_data
38
+ args: ckb
39
+ metrics:
40
+ - name: Test WER
41
+ type: wer
42
+ value: 0.36754389884276845
43
+ - name: Test CER
44
+ type: cer
45
+ value: 0.07827896768334217
46
+ ---
47
+
48
+ # Akashpb13/xlsr_hungarian_new
49
+
50
+ This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - hu dataset.
51
+ It achieves the following results on evaluation set (which is 10 percent of train data set merged with invalidated data, reported, other and dev datasets):
52
+ - Loss: 0.348580
53
+ - Wer: 0.401147
54
+
55
+ ## Model description
56
+ "facebook/wav2vec2-xls-r-300m" was finetuned.
57
+
58
+ ## Intended uses & limitations
59
+ More information needed
60
+ ## Training and evaluation data
61
+ Training data -
62
+ Common voice Central Kurdish train.tsv, dev.tsv, invalidated.tsv, reported.tsv, and other.tsv
63
+ Only those points were considered where upvotes were greater than downvotes and duplicates were removed after concatenation of all the datasets given in common voice 7.0
64
+
65
+ ## Training procedure
66
+ For creating the train dataset, all possible datasets were appended and 90-10 split was used.
67
+
68
+ ### Training hyperparameters
69
+
70
+ The following hyperparameters were used during training:
71
+
72
+ - learning_rate: 0.000095637994662983496
73
+ - train_batch_size: 16
74
+ - eval_batch_size: 16
75
+ - seed: 13
76
+ - gradient_accumulation_steps: 2
77
+ - lr_scheduler_type: cosine_with_restarts
78
+ - lr_scheduler_warmup_steps: 200
79
+ - num_epochs: 100
80
+ - mixed_precision_training: Native AMP
81
+
82
+
83
+ ### Training results
84
+
85
+ | Step | Training Loss | Validation Loss | Wer |
86
+ |-------|---------------|-----------------|----------|
87
+ | 500 | 5.097800 | 2.190326 | 1.001207 |
88
+ | 1000 | 0.797500 | 0.331392 | 0.576819 |
89
+ | 1500 | 0.405100 | 0.262009 | 0.549049 |
90
+ | 2000 | 0.322100 | 0.248178 | 0.479626 |
91
+ | 2500 | 0.264600 | 0.258866 | 0.488983 |
92
+ | 3000 | 0.228300 | 0.261523 | 0.469665 |
93
+ | 3500 | 0.201000 | 0.270135 | 0.451856 |
94
+ | 4000 | 0.180900 | 0.279302 | 0.448536 |
95
+ | 4500 | 0.163800 | 0.280921 | 0.459704 |
96
+ | 5000 | 0.147300 | 0.319249 | 0.471778 |
97
+ | 5500 | 0.137600 | 0.289546 | 0.449140 |
98
+ | 6000 | 0.132000 | 0.311350 | 0.458195 |
99
+ | 6500 | 0.117100 | 0.316726 | 0.432840 |
100
+ | 7000 | 0.109200 | 0.302210 | 0.439481 |
101
+ | 7500 | 0.104900 | 0.325913 | 0.439481 |
102
+ | 8000 | 0.097500 | 0.329446 | 0.431935 |
103
+ | 8500 | 0.088600 | 0.345259 | 0.425898 |
104
+ | 9000 | 0.084900 | 0.342891 | 0.428313 |
105
+ | 9500 | 0.080900 | 0.353081 | 0.424389 |
106
+ | 10000 | 0.075600 | 0.347063 | 0.424992 |
107
+ | 10500 | 0.072800 | 0.330086 | 0.424691 |
108
+ | 11000 | 0.068100 | 0.350658 | 0.421974 |
109
+ | 11500 | 0.064700 | 0.342949 | 0.413522 |
110
+ | 12000 | 0.061500 | 0.341704 | 0.415334 |
111
+ | 12500 | 0.059500 | 0.346279 | 0.411410 |
112
+ | 13000 | 0.057400 | 0.349901 | 0.407184 |
113
+ | 13500 | 0.056400 | 0.347733 | 0.402656 |
114
+ | 14000 | 0.053300 | 0.344899 | 0.405976 |
115
+ | 14500 | 0.052900 | 0.346708 | 0.402656 |
116
+ | 15000 | 0.050600 | 0.344118 | 0.400845 |
117
+ | 15500 | 0.050200 | 0.348396 | 0.402958 |
118
+ | 16000 | 0.049800 | 0.348312 | 0.401751 |
119
+ | 16500 | 0.051900 | 0.348372 | 0.401147 |
120
+ | 17000 | 0.049800 | 0.348580 | 0.401147 |
121
+
122
+
123
+
124
+ ### Framework versions
125
+ - Transformers 4.16.0.dev0
126
+ - Pytorch 1.10.0+cu102
127
+ - Datasets 1.17.1.dev0
128
+ - Tokenizers 0.10.3
129
+
130
+ #### Evaluation Commands
131
+
132
+ 1. To evaluate on `mozilla-foundation/common_voice_8_0` with split `test`
133
+
134
+ ```bash
135
+ python eval.py --model_id Akashpb13/Central_kurdish_xlsr --dataset mozilla-foundation/common_voice_8_0 --config ckb --split test
136
+ ```
137
+