Hugging Face's logo

deepdml
/
whisper-tiny-ar-mix-norm

Automatic Speech Recognition
Transformers
TensorBoard
Safetensors
Arabic
whisper
Generated from Trainer
Eval Results (legacy)
Model card Files
xet
Metrics Community
deepdml commited on
Commit
225dd67
·
verified ·
1 Parent(s): 241e55a

End of training

Browse files
Files changed (1) hide show
  1. README.md +28 -39
README.md CHANGED
@@ -7,12 +7,13 @@ base_model: openai/whisper-tiny
7
  tags:
8
  - generated_from_trainer
9
  datasets:
10
- - deepdml/Tunisian_MSA
11
- - fixie-ai/common_voice_17_0
12
  - pain/MASC
13
  - google/fleurs
14
- - UBC-NLP/Casablanca
 
15
  - ymoslem/MediaSpeech
 
 
16
  metrics:
17
  - wer
18
  model-index:
@@ -23,12 +24,13 @@ model-index:
23
  type: automatic-speech-recognition
24
  dataset:
25
  name: Common Voice 17.0
26
- type: deepdml/Tunisian_MSA
27
  metrics:
28
  - name: Wer
29
  type: wer
30
- value: 51.49189328143075
31
  ---
 
32
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
33
  should probably proofread and complete it, then remove this comment. -->
34
 
@@ -36,9 +38,9 @@ should probably proofread and complete it, then remove this comment. -->
36
 
37
  This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 17.0 dataset.
38
  It achieves the following results on the evaluation set:
39
- - Loss: 0.5877
40
- - Wer: 51.4919
41
- - Cer: 17.9569
42
 
43
  ## Model description
44
 
@@ -70,24 +72,24 @@ The following hyperparameters were used during training:
70
 
71
  | Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
72
  |:-------------:|:------:|:-----:|:---------------:|:-------:|:-------:|
73
- | 1.009 | 0.0556 | 1000 | 0.8091 | 68.4159 | 26.2902 |
74
- | 0.7599 | 0.1111 | 2000 | 0.7226 | 62.5420 | 23.2768 |
75
- | 0.562 | 0.1667 | 3000 | 0.6874 | 59.7455 | 21.5321 |
76
- | 0.4697 | 0.2222 | 4000 | 0.6690 | 57.8891 | 20.7745 |
77
- | 0.3596 | 0.2778 | 5000 | 0.6601 | 58.0562 | 21.0386 |
78
- | 0.3156 | 0.3333 | 6000 | 0.6483 | 56.2366 | 20.0099 |
79
- | 0.2326 | 0.3889 | 7000 | 0.6359 | 55.6215 | 19.8258 |
80
- | 0.193 | 0.4444 | 8000 | 0.6243 | 55.0118 | 19.7842 |
81
- | 0.1844 | 0.5 | 9000 | 0.6174 | 55.0100 | 19.8557 |
82
- | 0.1742 | 0.5556 | 10000 | 0.6142 | 53.9744 | 19.0535 |
83
- | 0.1408 | 0.6111 | 11000 | 0.6009 | 53.3721 | 19.1199 |
84
- | 0.1175 | 0.6667 | 12000 | 0.6040 | 52.5679 | 18.4452 |
85
- | 0.1441 | 0.7222 | 13000 | 0.6043 | 53.1298 | 18.8023 |
86
- | 0.1245 | 0.7778 | 14000 | 0.5880 | 52.0519 | 18.2682 |
87
- | 0.1189 | 0.8333 | 15000 | 0.5865 | 51.5103 | 18.1251 |
88
- | 0.1065 | 0.8889 | 16000 | 0.5863 | 51.3964 | 17.9292 |
89
- | 0.1266 | 0.9444 | 17000 | 0.5839 | 51.2275 | 17.9617 |
90
- | 0.1145 | 1.0 | 18000 | 0.5877 | 51.4919 | 17.9569 |
91
 
92
 
93
  ### Framework versions
@@ -96,16 +98,3 @@ The following hyperparameters were used during training:
96
  - Pytorch 2.5.1+cu121
97
  - Datasets 3.6.0
98
  - Tokenizers 0.21.0
99
-
100
- ## Citation
101
-
102
- Please cite the model using the following BibTeX entry:
103
-
104
- ```bibtex
105
- @misc{deepdml/whisper-tiny-ar-mix-norm,
106
- title={Fine-tuned Whisper tiny ASR model for speech recognition in Arabic},
107
- author={Jimenez, David},
108
- howpublished={\url{https://huggingface.co/deepdml/whisper-tiny-ar-mix-norm}},
109
- year={2026}
110
- }
111
- ```
 
7
  tags:
8
  - generated_from_trainer
9
  datasets:
 
 
10
  - pain/MASC
11
  - google/fleurs
12
+ - deepdml/Tunisian_MSA
13
+ - deepdml/mtedx
14
  - ymoslem/MediaSpeech
15
+ - UBC-NLP/Casablanca
16
+ - fixie-ai/common_voice_17_0
17
  metrics:
18
  - wer
19
  model-index:
 
24
  type: automatic-speech-recognition
25
  dataset:
26
  name: Common Voice 17.0
27
+ type: pain/MASC
28
  metrics:
29
  - name: Wer
30
  type: wer
31
+ value: 52.36224086961312
32
  ---
33
+
34
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
35
  should probably proofread and complete it, then remove this comment. -->
36
 
 
38
 
39
  This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 17.0 dataset.
40
  It achieves the following results on the evaluation set:
41
+ - Loss: 0.5962
42
+ - Wer: 52.3622
43
+ - Cer: 18.7245
44
 
45
  ## Model description
46
 
 
72
 
73
  | Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
74
  |:-------------:|:------:|:-----:|:---------------:|:-------:|:-------:|
75
+ | 1.1234 | 0.0556 | 1000 | 0.8106 | 67.0994 | 25.5658 |
76
+ | 0.8957 | 0.1111 | 2000 | 0.7234 | 62.5310 | 23.3911 |
77
+ | 0.7114 | 0.1667 | 3000 | 0.6871 | 59.7675 | 21.8722 |
78
+ | 0.6346 | 0.2222 | 4000 | 0.6637 | 58.1976 | 20.7336 |
79
+ | 0.4961 | 0.2778 | 5000 | 0.6545 | 57.7404 | 20.7048 |
80
+ | 0.4354 | 0.3333 | 6000 | 0.6473 | 56.7948 | 20.2061 |
81
+ | 0.3924 | 0.3889 | 7000 | 0.6325 | 55.8400 | 20.0139 |
82
+ | 0.3466 | 0.4444 | 8000 | 0.6274 | 55.4176 | 20.1441 |
83
+ | 0.2979 | 0.5 | 9000 | 0.6206 | 54.6997 | 19.6005 |
84
+ | 0.3099 | 0.5556 | 10000 | 0.6150 | 54.0166 | 19.3231 |
85
+ | 0.2681 | 0.6111 | 11000 | 0.6120 | 53.5980 | 19.1106 |
86
+ | 0.2383 | 0.6667 | 12000 | 0.6113 | 53.5576 | 19.4238 |
87
+ | 0.2582 | 0.7222 | 13000 | 0.6060 | 52.7515 | 18.7573 |
88
+ | 0.1543 | 0.7778 | 14000 | 0.6018 | 52.6175 | 18.4895 |
89
+ | 0.2356 | 0.8333 | 15000 | 0.6023 | 52.9902 | 18.9782 |
90
+ | 0.2031 | 0.8889 | 16000 | 0.5984 | 52.5165 | 18.8550 |
91
+ | 0.2437 | 0.9444 | 17000 | 0.5951 | 52.4926 | 18.7514 |
92
+ | 0.2269 | 1.0 | 18000 | 0.5962 | 52.3622 | 18.7245 |
93
 
94
 
95
  ### Framework versions
 
98
  - Pytorch 2.5.1+cu121
99
  - Datasets 3.6.0
100
  - Tokenizers 0.21.0