Hugging Face's logo

golesheed
/
wav2vec2-large-xls-r-1b-frisian-cv-13-elderly-augmented

Automatic Speech Recognition
Transformers
PyTorch
TensorBoard
wav2vec2
Generated from Trainer
Model card Files
xet
 Metrics Community
Golshid Shekoufandeh commited on
Commit
eecc18f
·
1 Parent(s): 2c5bef1

update model card README.md

Browse files
Files changed (1) hide show
  1. README.md +15 -31
README.md CHANGED
@@ -16,8 +16,8 @@ should probably proofread and complete it, then remove this comment. -->
16
 
17
  This model is a fine-tuned version of [greenw0lf/wav2vec2-large-xls-r-1b-frisian](https://huggingface.co/greenw0lf/wav2vec2-large-xls-r-1b-frisian) on an unknown dataset.
18
  It achieves the following results on the evaluation set:
19
- - Loss: 0.1970
20
- - Wer: 0.2588
21
 
22
  ## Model description
23
 
@@ -43,41 +43,25 @@ The following hyperparameters were used during training:
43
  - optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
44
  - lr_scheduler_type: linear
45
  - lr_scheduler_warmup_ratio: 0.1
46
- - num_epochs: 70
47
  - mixed_precision_training: Native AMP
48
 
49
  ### Training results
50
 
51
  | Training Loss | Epoch | Step | Validation Loss | Wer |
52
  |:-------------:|:-----:|:----:|:---------------:|:------:|
53
- | 11.3951 | 2.5 | 50 | 6.0419 | 1.0 |
54
- | 5.6022 | 5.0 | 100 | 3.8553 | 1.0009 |
55
- | 4.2114 | 7.5 | 150 | 3.5111 | 1.0013 |
56
- | 3.862 | 10.0 | 200 | 3.3501 | 1.0009 |
57
- | 3.4295 | 12.5 | 250 | 3.2998 | 1.0 |
58
- | 3.0977 | 15.0 | 300 | 3.2271 | 1.0 |
59
- | 3.0654 | 17.5 | 350 | 3.1985 | 0.9915 |
60
- | 2.9113 | 20.0 | 400 | 3.1006 | 0.9928 |
61
- | 2.3635 | 22.5 | 450 | 0.7427 | 0.6536 |
62
- | 1.1831 | 25.0 | 500 | 0.2535 | 0.2973 |
63
- | 0.807 | 27.5 | 550 | 0.2128 | 0.2816 |
64
- | 0.8789 | 30.0 | 600 | 0.1943 | 0.2664 |
65
- | 0.634 | 32.5 | 650 | 0.1887 | 0.2597 |
66
- | 0.4828 | 35.0 | 700 | 0.1946 | 0.2584 |
67
- | 0.4695 | 37.5 | 750 | 0.1944 | 0.2637 |
68
- | 0.7161 | 40.0 | 800 | 0.1944 | 0.2597 |
69
- | 0.7334 | 42.5 | 850 | 0.1944 | 0.2606 |
70
- | 0.5272 | 45.0 | 900 | 0.1990 | 0.2597 |
71
- | 0.6171 | 47.5 | 950 | 0.1984 | 0.2620 |
72
- | 0.4454 | 50.0 | 1000 | 0.2010 | 0.2606 |
73
- | 0.6424 | 52.5 | 1050 | 0.2000 | 0.2642 |
74
- | 0.8542 | 55.0 | 1100 | 0.1962 | 0.2642 |
75
- | 0.5866 | 57.5 | 1150 | 0.1971 | 0.2611 |
76
- | 0.4219 | 60.0 | 1200 | 0.1965 | 0.2593 |
77
- | 0.3456 | 62.5 | 1250 | 0.1967 | 0.2588 |
78
- | 0.3567 | 65.0 | 1300 | 0.1962 | 0.2597 |
79
- | 0.4524 | 67.5 | 1350 | 0.1977 | 0.2588 |
80
- | 0.4635 | 70.0 | 1400 | 0.1970 | 0.2588 |
81
 
82
 
83
  ### Framework versions
 
16
 
17
  This model is a fine-tuned version of [greenw0lf/wav2vec2-large-xls-r-1b-frisian](https://huggingface.co/greenw0lf/wav2vec2-large-xls-r-1b-frisian) on an unknown dataset.
18
  It achieves the following results on the evaluation set:
19
+ - Loss: 0.1937
20
+ - Wer: 0.2602
21
 
22
  ## Model description
23
 
 
43
  - optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
44
  - lr_scheduler_type: linear
45
  - lr_scheduler_warmup_ratio: 0.1
46
+ - num_epochs: 30
47
  - mixed_precision_training: Native AMP
48
 
49
  ### Training results
50
 
51
  | Training Loss | Epoch | Step | Validation Loss | Wer |
52
  |:-------------:|:-----:|:----:|:---------------:|:------:|
53
+ | 8.7364 | 2.5 | 50 | 2.9854 | 1.0 |
54
+ | 1.9624 | 5.0 | 100 | 0.3770 | 0.2812 |
55
+ | 0.8855 | 7.5 | 150 | 0.2479 | 0.2499 |
56
+ | 0.5855 | 10.0 | 200 | 0.2291 | 0.2481 |
57
+ | 0.44 | 12.5 | 250 | 0.2096 | 0.2490 |
58
+ | 0.7951 | 15.0 | 300 | 0.1971 | 0.2553 |
59
+ | 0.7498 | 17.5 | 350 | 0.1958 | 0.2521 |
60
+ | 0.3246 | 20.0 | 400 | 0.1963 | 0.2651 |
61
+ | 0.4627 | 22.5 | 450 | 0.1925 | 0.2588 |
62
+ | 0.4105 | 25.0 | 500 | 0.1922 | 0.2611 |
63
+ | 0.5051 | 27.5 | 550 | 0.1949 | 0.2593 |
64
+ | 0.6736 | 30.0 | 600 | 0.1937 | 0.2602 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
65
 
66
 
67
  ### Framework versions