saattrupdan
commited on
Commit
•
1858418
1
Parent(s):
a6e8685
Update README.md
Browse files
README.md
CHANGED
@@ -1,127 +1,26 @@
|
|
1 |
---
|
|
|
|
|
2 |
license: cc0-1.0
|
3 |
-
tags:
|
4 |
-
- generated_from_trainer
|
5 |
datasets:
|
6 |
-
-
|
7 |
model-index:
|
8 |
- name: kblab-voxrex-wav2vec2-large-cv8-da
|
9 |
results: []
|
10 |
---
|
11 |
|
12 |
-
|
13 |
-
should probably proofread and complete it, then remove this comment. -->
|
14 |
-
|
15 |
-
# kblab-voxrex-wav2vec2-large-cv8-da
|
16 |
-
|
17 |
-
This model is a fine-tuned version of [KBLab/wav2vec2-large-voxrex](https://huggingface.co/KBLab/wav2vec2-large-voxrex) on the common_voice dataset.
|
18 |
-
It achieves the following results on the evaluation set:
|
19 |
-
- Loss: 329.0055
|
20 |
-
- Wer: 0.3768
|
21 |
|
22 |
## Model description
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
## Intended uses & limitations
|
27 |
-
|
28 |
-
More information needed
|
29 |
-
|
30 |
-
## Training and evaluation data
|
31 |
-
|
32 |
-
More information needed
|
33 |
-
|
34 |
-
## Training procedure
|
35 |
-
|
36 |
-
### Training hyperparameters
|
37 |
-
|
38 |
-
The following hyperparameters were used during training:
|
39 |
-
- learning_rate: 4e-05
|
40 |
-
- train_batch_size: 4
|
41 |
-
- eval_batch_size: 8
|
42 |
-
- seed: 4242
|
43 |
-
- gradient_accumulation_steps: 8
|
44 |
-
- total_train_batch_size: 32
|
45 |
-
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
46 |
-
- lr_scheduler_type: linear
|
47 |
-
- lr_scheduler_warmup_steps: 500
|
48 |
-
- num_epochs: 500
|
49 |
-
- mixed_precision_training: Native AMP
|
50 |
-
|
51 |
-
### Training results
|
52 |
|
53 |
-
| Training Loss | Epoch | Step | Validation Loss | Wer |
|
54 |
-
|:-------------:|:------:|:-----:|:---------------:|:------:|
|
55 |
-
| 767.4506 | 5.55 | 300 | 1359.1575 | 1.0 |
|
56 |
-
| 576.4063 | 11.11 | 600 | 1265.8390 | 1.0 |
|
57 |
-
| 519.8654 | 16.66 | 900 | 1039.6812 | 1.0 |
|
58 |
-
| 313.227 | 22.22 | 1200 | 551.1237 | 0.8551 |
|
59 |
-
| 241.8147 | 27.77 | 1500 | 421.1035 | 0.7096 |
|
60 |
-
| 203.6478 | 33.33 | 1800 | 359.2018 | 0.6291 |
|
61 |
-
| 169.5277 | 38.88 | 2100 | 328.7173 | 0.5931 |
|
62 |
-
| 149.7277 | 44.44 | 2400 | 312.2329 | 0.5593 |
|
63 |
-
| 134.0794 | 49.99 | 2700 | 298.8540 | 0.5364 |
|
64 |
-
| 124.439 | 55.55 | 3000 | 295.4873 | 0.5169 |
|
65 |
-
| 114.4032 | 61.11 | 3300 | 287.1676 | 0.5050 |
|
66 |
-
| 103.9973 | 66.66 | 3600 | 280.2365 | 0.4967 |
|
67 |
-
| 96.152 | 72.22 | 3900 | 279.2440 | 0.4857 |
|
68 |
-
| 89.5619 | 77.77 | 4200 | 279.0049 | 0.4739 |
|
69 |
-
| 89.8041 | 83.33 | 4500 | 276.0360 | 0.4616 |
|
70 |
-
| 78.6993 | 88.88 | 4800 | 278.6253 | 0.4539 |
|
71 |
-
| 74.2165 | 94.44 | 5100 | 276.4348 | 0.4488 |
|
72 |
-
| 69.5902 | 99.99 | 5400 | 276.1476 | 0.4417 |
|
73 |
-
| 67.8592 | 105.55 | 5700 | 275.3440 | 0.4341 |
|
74 |
-
| 64.1541 | 111.11 | 6000 | 278.0880 | 0.4363 |
|
75 |
-
| 60.7204 | 116.66 | 6300 | 281.5571 | 0.4374 |
|
76 |
-
| 56.6715 | 122.22 | 6600 | 282.7102 | 0.4306 |
|
77 |
-
| 55.7875 | 127.77 | 6900 | 279.3789 | 0.4228 |
|
78 |
-
| 54.5305 | 133.33 | 7200 | 283.6728 | 0.4208 |
|
79 |
-
| 51.4744 | 138.88 | 7500 | 282.4348 | 0.4227 |
|
80 |
-
| 47.1217 | 144.44 | 7800 | 287.4393 | 0.4123 |
|
81 |
-
| 48.4808 | 149.99 | 8100 | 286.8406 | 0.4126 |
|
82 |
-
| 46.415 | 155.55 | 8400 | 290.3094 | 0.4144 |
|
83 |
-
| 43.29 | 161.11 | 8700 | 291.6872 | 0.4144 |
|
84 |
-
| 42.7431 | 166.66 | 9000 | 297.7512 | 0.4210 |
|
85 |
-
| 41.8859 | 172.22 | 9300 | 296.6982 | 0.4085 |
|
86 |
-
| 41.2126 | 177.77 | 9600 | 294.0860 | 0.4123 |
|
87 |
-
| 40.8457 | 183.33 | 9900 | 298.7288 | 0.4058 |
|
88 |
-
| 36.6865 | 188.88 | 10200 | 305.0593 | 0.4036 |
|
89 |
-
| 34.1681 | 194.44 | 10500 | 304.9405 | 0.4112 |
|
90 |
-
| 34.4368 | 199.99 | 10800 | 303.7193 | 0.4023 |
|
91 |
-
| 35.3407 | 205.55 | 11100 | 295.9553 | 0.3975 |
|
92 |
-
| 34.0598 | 211.11 | 11400 | 300.0461 | 0.4012 |
|
93 |
-
| 33.4694 | 216.66 | 11700 | 307.4055 | 0.3942 |
|
94 |
-
| 32.2768 | 222.22 | 12000 | 307.5330 | 0.3926 |
|
95 |
-
| 34.4758 | 227.77 | 12300 | 307.9725 | 0.4003 |
|
96 |
-
| 30.5966 | 233.33 | 12600 | 311.4758 | 0.3950 |
|
97 |
-
| 29.2803 | 238.88 | 12900 | 308.0916 | 0.3933 |
|
98 |
-
| 28.6945 | 244.44 | 13200 | 307.3855 | 0.3921 |
|
99 |
-
| 29.8094 | 249.99 | 13500 | 317.3207 | 0.3920 |
|
100 |
-
| 29.7135 | 255.55 | 13800 | 310.4784 | 0.3925 |
|
101 |
-
| 28.7815 | 261.11 | 14100 | 315.4926 | 0.3942 |
|
102 |
-
| 27.1585 | 266.66 | 14400 | 321.6101 | 0.3972 |
|
103 |
-
| 26.9533 | 272.22 | 14700 | 314.2688 | 0.3918 |
|
104 |
-
| 26.8752 | 277.77 | 15000 | 321.5280 | 0.3941 |
|
105 |
-
| 26.7076 | 283.33 | 15300 | 323.5451 | 0.3912 |
|
106 |
-
| 25.8936 | 288.88 | 15600 | 326.1316 | 0.3889 |
|
107 |
-
| 25.6714 | 294.44 | 15900 | 324.0426 | 0.3905 |
|
108 |
-
| 25.0952 | 299.99 | 16200 | 322.3788 | 0.3870 |
|
109 |
-
| 23.5694 | 305.55 | 16500 | 323.4653 | 0.3828 |
|
110 |
-
| 24.6763 | 311.11 | 16800 | 328.5225 | 0.3831 |
|
111 |
-
| 25.1798 | 316.66 | 17100 | 320.6808 | 0.3868 |
|
112 |
-
| 23.6551 | 322.22 | 17400 | 325.5733 | 0.3842 |
|
113 |
-
| 23.118 | 327.77 | 17700 | 327.0573 | 0.3820 |
|
114 |
-
| 23.178 | 333.33 | 18000 | 322.2932 | 0.3723 |
|
115 |
-
| 22.3727 | 338.88 | 18300 | 332.8637 | 0.3783 |
|
116 |
-
| 22.8178 | 344.44 | 18600 | 333.7156 | 0.3854 |
|
117 |
-
| 22.3476 | 349.99 | 18900 | 326.8071 | 0.3766 |
|
118 |
-
| 21.6792 | 355.55 | 19200 | 329.8040 | 0.3793 |
|
119 |
-
| 23.5751 | 361.11 | 19500 | 329.0055 | 0.3768 |
|
120 |
|
|
|
121 |
|
122 |
-
|
123 |
|
124 |
-
-
|
125 |
-
|
126 |
-
-
|
127 |
-
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- da
|
4 |
license: cc0-1.0
|
|
|
|
|
5 |
datasets:
|
6 |
+
- common_voice_8_0
|
7 |
model-index:
|
8 |
- name: kblab-voxrex-wav2vec2-large-cv8-da
|
9 |
results: []
|
10 |
---
|
11 |
|
12 |
+
# KBLab-VoxRex-Wav2vec2-large-CV8-da
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
## Model description
|
15 |
|
16 |
+
This model is a fine-tuned version of the Swedish acoustic model [KBLab/wav2vec2-large-voxrex](https://huggingface.co/KBLab/wav2vec2-large-voxrex) on the Danish part of [Common Voice 8.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0), containing ~6 crowdsourced hours of read-aloud Danish speech.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
+
## Performance
|
20 |
|
21 |
+
The model achieves the following WER scores (lower is better):
|
22 |
|
23 |
+
| **Dataset** | **WER without LM** | **WER with 5-gram LM** |
|
24 |
+
| :---: | ---: | ---: |
|
25 |
+
| [Danish part of Common Voice 8.0](https://huggingface.co/datasets/mozilla-foundation/common_voice_8_0/viewer/da/train) | 37.63 | xx.xx |
|
26 |
+
| [Alvenir test set](https://huggingface.co/datasets/Alvenir/alvenir_asr_da_eval) | 35.75 | xx.xx |
|