Hugging Face's logo

Akashpb13
/
Kabyle_xlsr

Automatic Speech Recognition
Transformers
PyTorch
Safetensors
Kabyle
wav2vec2
mozilla-foundation/common_voice_8_0
Generated from Trainer
sw
robust-speech-event
model_for_talk
hf-asr-leaderboard
Eval Results (legacy)
Model card Files
xet
 Community
3
Akashpb13 commited on
Commit
f1e1d8f
·
1 Parent(s): a16635c

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +165 -0
README.md ADDED
@@ -0,0 +1,165 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - kab
4
+ license: apache-2.0
5
+ tags:
6
+ - automatic-speech-recognition
7
+ - mozilla-foundation/common_voice_8_0
8
+ - generated_from_trainer
9
+ - sw
10
+ - robust-speech-event
11
+ - model_for_talk
12
+ datasets:
13
+ - mozilla-foundation/common_voice_8_0
14
+
15
+ model-index:
16
+ - name: Akashpb13/Kabyle_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: kab
25
+ metrics:
26
+ - name: Test WER
27
+ type: wer
28
+ value: 0.3188425282720088
29
+ - name: Test CER
30
+ type: cer
31
+ value: 0.09443079928558358
32
+
33
+
34
+ ---
35
+
36
+ # Akashpb13/xlsr_hungarian_new
37
+
38
+ 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.
39
+ It achieves the following results on the evaluation set (which is 10 percent of train data set merged with dev datasets):
40
+ - Loss: 0.159032
41
+ - Wer: 0.187934
42
+ ## Model description
43
+ "facebook/wav2vec2-xls-r-300m" was finetuned.
44
+
45
+ ## Intended uses & limitations
46
+ More information needed
47
+ ## Training and evaluation data
48
+ Training data -
49
+ Common voice Kabyle train.tsv. Only 50,000 records were sampled randomly and trained due to huge size of dataset.
50
+ 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
51
+
52
+ ## Training procedure
53
+ For creating the training dataset, all possible datasets were appended and 90-10 split was used.
54
+
55
+ ### Training hyperparameters
56
+
57
+ The following hyperparameters were used during training:
58
+
59
+ - learning_rate: 0.000096
60
+ - train_batch_size: 8
61
+ - seed: 13
62
+ - gradient_accumulation_steps: 4
63
+ - lr_scheduler_type: cosine_with_restarts
64
+ - lr_scheduler_warmup_steps: 500
65
+ - num_epochs: 30
66
+ - mixed_precision_training: Native AMP
67
+
68
+
69
+ ### Training results
70
+ | Step | Training Loss | Validation Loss | Wer |
71
+ |-------|---------------|-----------------|----------|
72
+ | 500 | 7.199800 | 3.130564 | 1.000000 |
73
+ | 1000 | 1.570200 | 0.718097 | 0.734682 |
74
+ | 1500 | 0.850800 | 0.524227 | 0.640532 |
75
+ | 2000 | 0.712200 | 0.468694 | 0.603454 |
76
+ | 2500 | 0.651200 | 0.413833 | 0.573025 |
77
+ | 3000 | 0.603100 | 0.403680 | 0.552847 |
78
+ | 3500 | 0.553300 | 0.372638 | 0.541719 |
79
+ | 4000 | 0.537200 | 0.353759 | 0.531191 |
80
+ | 4500 | 0.506300 | 0.359109 | 0.519601 |
81
+ | 5000 | 0.479600 | 0.343937 | 0.511336 |
82
+ | 5500 | 0.479800 | 0.338214 | 0.503948 |
83
+ | 6000 | 0.449500 | 0.332600 | 0.495221 |
84
+ | 6500 | 0.439200 | 0.323905 | 0.492635 |
85
+ | 7000 | 0.434900 | 0.310417 | 0.484555 |
86
+ | 7500 | 0.403200 | 0.311247 | 0.483262 |
87
+ | 8000 | 0.401500 | 0.295637 | 0.476566 |
88
+ | 8500 | 0.397000 | 0.301321 | 0.471672 |
89
+ | 9000 | 0.371600 | 0.295639 | 0.468440 |
90
+ | 9500 | 0.370700 | 0.294039 | 0.468902 |
91
+ | 10000 | 0.364900 | 0.291195 | 0.468440 |
92
+ | 10500 | 0.348300 | 0.284898 | 0.461098 |
93
+ | 11000 | 0.350100 | 0.281764 | 0.459805 |
94
+ | 11500 | 0.336900 | 0.291022 | 0.461606 |
95
+ | 12000 | 0.330700 | 0.280467 | 0.455234 |
96
+ | 12500 | 0.322500 | 0.271714 | 0.452694 |
97
+ | 13000 | 0.307400 | 0.289519 | 0.455465 |
98
+ | 13500 | 0.309300 | 0.281922 | 0.451217 |
99
+ | 14000 | 0.304800 | 0.271514 | 0.452186 |
100
+ | 14500 | 0.288100 | 0.286801 | 0.446830 |
101
+ | 15000 | 0.293200 | 0.276309 | 0.445399 |
102
+ | 15500 | 0.289800 | 0.287188 | 0.446230 |
103
+ | 16000 | 0.274800 | 0.286406 | 0.441243 |
104
+ | 16500 | 0.271700 | 0.284754 | 0.441520 |
105
+ | 17000 | 0.262500 | 0.275431 | 0.442167 |
106
+ | 17500 | 0.255500 | 0.276575 | 0.439858 |
107
+ | 18000 | 0.260200 | 0.269911 | 0.435425 |
108
+ | 18500 | 0.250600 | 0.270519 | 0.434686 |
109
+ | 19000 | 0.243300 | 0.267655 | 0.437826 |
110
+ | 19500 | 0.240600 | 0.277109 | 0.431731 |
111
+ | 20000 | 0.237200 | 0.266622 | 0.433994 |
112
+ | 20500 | 0.231300 | 0.273015 | 0.428868 |
113
+ | 21000 | 0.227200 | 0.263024 | 0.430161 |
114
+ | 21500 | 0.220400 | 0.272880 | 0.429607 |
115
+ | 22000 | 0.218600 | 0.272340 | 0.426883 |
116
+ | 22500 | 0.213100 | 0.277066 | 0.428407 |
117
+ | 23000 | 0.205000 | 0.278404 | 0.424020 |
118
+ | 23500 | 0.200900 | 0.270877 | 0.418987 |
119
+ | 24000 | 0.199000 | 0.289120 | 0.425821 |
120
+ | 24500 | 0.196100 | 0.275831 | 0.424066 |
121
+ | 25000 | 0.191100 | 0.282822 | 0.421850 |
122
+ | 25500 | 0.190100 | 0.275820 | 0.418248 |
123
+ | 26000 | 0.178800 | 0.279208 | 0.419125 |
124
+ | 26500 | 0.183100 | 0.271464 | 0.419218 |
125
+ | 27000 | 0.177400 | 0.280869 | 0.419680 |
126
+ | 27500 | 0.171800 | 0.279593 | 0.414924 |
127
+ | 28000 | 0.172900 | 0.276949 | 0.417648 |
128
+ | 28500 | 0.164900 | 0.283491 | 0.417786 |
129
+ | 29000 | 0.164800 | 0.283122 | 0.416078 |
130
+ | 29500 | 0.165500 | 0.281969 | 0.415801 |
131
+ | 30000 | 0.163800 | 0.283319 | 0.412753 |
132
+ | 30500 | 0.153500 | 0.285702 | 0.414046 |
133
+ | 31000 | 0.156500 | 0.285041 | 0.412615 |
134
+ | 31500 | 0.150900 | 0.284336 | 0.413723 |
135
+ | 32000 | 0.151800 | 0.285922 | 0.412292 |
136
+ | 32500 | 0.149200 | 0.289461 | 0.412153 |
137
+ | 33000 | 0.145400 | 0.291322 | 0.409567 |
138
+ | 33500 | 0.145600 | 0.294361 | 0.409614 |
139
+ | 34000 | 0.144200 | 0.290686 | 0.409059 |
140
+ | 34500 | 0.143400 | 0.289474 | 0.409844 |
141
+ | 35000 | 0.143500 | 0.290340 | 0.408367 |
142
+ | 35500 | 0.143200 | 0.289581 | 0.407351 |
143
+ | 36000 | 0.138400 | 0.292782 | 0.408736 |
144
+ | 36500 | 0.137900 | 0.289108 | 0.408044 |
145
+ | 37000 | 0.138200 | 0.292127 | 0.407166 |
146
+ | 37500 | 0.134600 | 0.291797 | 0.408413 |
147
+ | 38000 | 0.139800 | 0.290056 | 0.408090 |
148
+ | 38500 | 0.136500 | 0.291198 | 0.408090 |
149
+ | 39000 | 0.137700 | 0.289696 | 0.408044 |
150
+
151
+
152
+ ### Framework versions
153
+ - Transformers 4.16.0.dev0
154
+ - Pytorch 1.10.0+cu102
155
+ - Datasets 1.18.3
156
+ - Tokenizers 0.10.3
157
+
158
+ #### Evaluation Commands
159
+
160
+ 1. To evaluate on `mozilla-foundation/common_voice_8_0` with split `test`
161
+
162
+ ```bash
163
+ python eval.py --model_id Akashpb13/Kabyle_xlsr --dataset mozilla-foundation/common_voice_8_0 --config kab --split test
164
+ ```
165
+