Hugging Face's logo

deepdml
/
whisper-tiny-af-mix-norm

TensorBoard
Safetensors
Afrikaans
whisper
Generated from Trainer
Eval Results (legacy)
Model card Files
xet
 Metrics Community
deepdml commited on
Commit
16dbc5e
·
1 Parent(s): 3adf8d9

End of training

Browse files
Files changed (1) hide show
  1. README.md +30 -40
README.md CHANGED
@@ -6,8 +6,10 @@ base_model: openai/whisper-tiny
6
  tags:
7
  - generated_from_trainer
8
  datasets:
 
9
  - voice-biomarkers/openslr-32-hq-SA-languages-Afrikaans
10
  - google/fleurs
 
11
  metrics:
12
  - wer
13
  model-index:
@@ -18,15 +20,16 @@ model-index:
18
  type: automatic-speech-recognition
19
  dataset:
20
  name: Common Voice 17.0
21
- type: voice-biomarkers/openslr-32-hq-SA-languages-Afrikaans
22
  config: af_za
23
  split: test
24
  args: af_za
25
  metrics:
26
  - name: Wer
27
  type: wer
28
- value: 52.17316017316017
29
  ---
 
30
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
31
  should probably proofread and complete it, then remove this comment. -->
32
 
@@ -34,9 +37,9 @@ should probably proofread and complete it, then remove this comment. -->
34
 
35
  This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 17.0 dataset.
36
  It achieves the following results on the evaluation set:
37
- - Loss: 1.3668
38
- - Wer: 52.1732
39
- - Cer: 20.9395
40
 
41
  ## Model description
42
 
@@ -66,28 +69,28 @@ The following hyperparameters were used during training:
66
 
67
  ### Training results
68
 
69
- | Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
70
- |:-------------:|:-------:|:----:|:---------------:|:-------:|:-------:|
71
- | 1.3199 | 0.05 | 100 | 1.4770 | 59.2208 | 24.5022 |
72
- | 0.6393 | 1.0315 | 200 | 1.2510 | 51.2554 | 20.9454 |
73
- | 0.3811 | 2.013 | 300 | 1.2197 | 49.4545 | 20.1155 |
74
- | 0.261 | 2.063 | 400 | 1.2089 | 48.3290 | 19.2036 |
75
- | 0.186 | 3.0445 | 500 | 1.2141 | 48.0693 | 19.8575 |
76
- | 0.1459 | 4.026 | 600 | 1.2341 | 49.8701 | 20.2621 |
77
- | 0.0963 | 5.0075 | 700 | 1.2517 | 48.4675 | 19.5437 |
78
- | 0.0809 | 5.0575 | 800 | 1.2674 | 51.0823 | 21.0715 |
79
- | 0.0536 | 6.039 | 900 | 1.2812 | 48.2597 | 19.5408 |
80
- | 0.0432 | 7.0205 | 1000 | 1.3003 | 48.5022 | 19.4910 |
81
- | 0.0379 | 8.002 | 1100 | 1.3117 | 51.6190 | 21.2298 |
82
- | 0.0333 | 8.052 | 1200 | 1.3314 | 52.3463 | 21.7078 |
83
- | 0.0247 | 9.0335 | 1300 | 1.3389 | 52.0 | 21.4644 |
84
- | 0.0201 | 10.015 | 1400 | 1.3484 | 51.4113 | 22.1769 |
85
- | 0.0194 | 10.065 | 1500 | 1.3469 | 51.8442 | 21.0685 |
86
- | 0.0191 | 11.0465 | 1600 | 1.3536 | 52.4502 | 21.3471 |
87
- | 0.0179 | 12.028 | 1700 | 1.3611 | 51.7229 | 21.1272 |
88
- | 0.0155 | 13.0095 | 1800 | 1.3637 | 52.4329 | 20.9512 |
89
- | 0.0159 | 13.0595 | 1900 | 1.3651 | 52.0346 | 20.9483 |
90
- | 0.0152 | 14.041 | 2000 | 1.3668 | 52.1732 | 20.9395 |
91
 
92
 
93
  ### Framework versions
@@ -96,16 +99,3 @@ The following hyperparameters were used during training:
96
  - Pytorch 2.3.0+cu121
97
  - Datasets 2.19.1
98
  - Tokenizers 0.19.1
99
-
100
- ## Citation
101
-
102
- Please cite the model using the following BibTeX entry:
103
-
104
- ```bibtex
105
- @misc{deepdml/whisper-tiny-af-mix-norm,
106
- title={Fine-tuned Whisper tiny ASR model for speech recognition in Afrikaans},
107
- author={Jimenez, David},
108
- howpublished={\url{https://huggingface.co/deepdml/whisper-tiny-af-mix-norm}},
109
- year={2026}
110
- }
111
- ```
 
6
  tags:
7
  - generated_from_trainer
8
  datasets:
9
+ - andreoosthuizen/afrikaans-30s
10
  - voice-biomarkers/openslr-32-hq-SA-languages-Afrikaans
11
  - google/fleurs
12
+ - dsfsi-anv/multilingual-nchlt-dataset
13
  metrics:
14
  - wer
15
  model-index:
 
20
  type: automatic-speech-recognition
21
  dataset:
22
  name: Common Voice 17.0
23
+ type: andreoosthuizen/afrikaans-30s
24
  config: af_za
25
  split: test
26
  args: af_za
27
  metrics:
28
  - name: Wer
29
  type: wer
30
+ value: 44.935064935064936
31
  ---
32
+
33
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
34
  should probably proofread and complete it, then remove this comment. -->
35
 
 
37
 
38
  This model is a fine-tuned version of [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) on the Common Voice 17.0 dataset.
39
  It achieves the following results on the evaluation set:
40
+ - Loss: 1.1668
41
+ - Wer: 44.9351
42
+ - Cer: 18.2741
43
 
44
  ## Model description
45
 
 
69
 
70
  ### Training results
71
 
72
+ | Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
73
+ |:-------------:|:-----:|:----:|:---------------:|:-------:|:-------:|
74
+ | 1.3743 | 0.05 | 100 | 1.5130 | 63.5498 | 26.0798 |
75
+ | 0.7811 | 0.1 | 200 | 1.2491 | 51.7403 | 19.9455 |
76
+ | 0.5477 | 0.15 | 300 | 1.1820 | 48.1732 | 18.7168 |
77
+ | 0.4289 | 0.2 | 400 | 1.1518 | 49.6277 | 19.5203 |
78
+ | 0.3573 | 0.25 | 500 | 1.1410 | 48.6234 | 19.9044 |
79
+ | 0.2835 | 0.3 | 600 | 1.1289 | 47.3074 | 19.3649 |
80
+ | 0.2602 | 0.35 | 700 | 1.1318 | 45.7835 | 19.3150 |
81
+ | 0.217 | 0.4 | 800 | 1.1297 | 46.8398 | 19.1361 |
82
+ | 0.2007 | 0.45 | 900 | 1.1358 | 47.4286 | 20.3296 |
83
+ | 0.1798 | 0.5 | 1000 | 1.1383 | 47.4459 | 20.5906 |
84
+ | 0.1548 | 0.55 | 1100 | 1.1497 | 49.3853 | 21.7723 |
85
+ | 0.1384 | 0.6 | 1200 | 1.1525 | 48.5022 | 20.2827 |
86
+ | 0.1325 | 0.65 | 1300 | 1.1574 | 48.8831 | 20.3120 |
87
+ | 0.1259 | 0.7 | 1400 | 1.1625 | 45.4372 | 18.7637 |
88
+ | 0.125 | 0.75 | 1500 | 1.1606 | 44.7100 | 18.1128 |
89
+ | 0.1083 | 0.8 | 1600 | 1.1609 | 48.0519 | 20.1507 |
90
+ | 0.1169 | 0.85 | 1700 | 1.1660 | 47.6017 | 19.6112 |
91
+ | 0.1008 | 0.9 | 1800 | 1.1644 | 47.8095 | 19.9924 |
92
+ | 0.1016 | 0.95 | 1900 | 1.1658 | 44.7273 | 18.1626 |
93
+ | 0.0983 | 1.0 | 2000 | 1.1668 | 44.9351 | 18.2741 |
94
 
95
 
96
  ### Framework versions
 
99
  - Pytorch 2.3.0+cu121
100
  - Datasets 2.19.1
101
  - Tokenizers 0.19.1