đ wav2vec2-large-xls-r-300m-spanish-custom
This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It can achieve high - quality speech recognition results, with a loss of 0.4426 and a word error rate (Wer) of 0.2117 on the evaluation set.
đ Quick Start
This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset.
It achieves the following results on the evaluation set:
đ§ Technical Details
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e - 08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
Training Loss |
Epoch |
Step |
Validation Loss |
Wer |
4.2307 |
0.4 |
400 |
1.4431 |
0.9299 |
0.7066 |
0.79 |
800 |
0.5928 |
0.4836 |
0.4397 |
1.19 |
1200 |
0.4341 |
0.3730 |
0.3889 |
1.58 |
1600 |
0.4063 |
0.3499 |
0.3607 |
1.98 |
2000 |
0.3834 |
0.3235 |
0.2866 |
2.37 |
2400 |
0.3885 |
0.3163 |
0.2833 |
2.77 |
2800 |
0.3765 |
0.3140 |
0.2692 |
3.17 |
3200 |
0.3849 |
0.3132 |
0.2435 |
3.56 |
3600 |
0.3779 |
0.2984 |
0.2404 |
3.96 |
4000 |
0.3756 |
0.2934 |
0.2153 |
4.35 |
4400 |
0.3770 |
0.3075 |
0.2087 |
4.75 |
4800 |
0.3819 |
0.3022 |
0.1999 |
5.14 |
5200 |
0.3756 |
0.2959 |
0.1838 |
5.54 |
5600 |
0.3827 |
0.2858 |
0.1892 |
5.93 |
6000 |
0.3714 |
0.2999 |
0.1655 |
6.33 |
6400 |
0.3814 |
0.2812 |
0.1649 |
6.73 |
6800 |
0.3685 |
0.2727 |
0.1668 |
7.12 |
7200 |
0.3832 |
0.2825 |
0.1487 |
7.52 |
7600 |
0.3848 |
0.2788 |
0.152 |
7.91 |
8000 |
0.3810 |
0.2787 |
0.143 |
8.31 |
8400 |
0.3885 |
0.2856 |
0.1353 |
8.7 |
8800 |
0.4103 |
0.2827 |
0.1386 |
9.1 |
9200 |
0.4142 |
0.2874 |
0.1222 |
9.5 |
9600 |
0.3983 |
0.2830 |
0.1288 |
9.89 |
10000 |
0.4179 |
0.2781 |
0.1199 |
10.29 |
10400 |
0.4035 |
0.2789 |
0.1196 |
10.68 |
10800 |
0.4043 |
0.2746 |
0.1169 |
11.08 |
11200 |
0.4105 |
0.2753 |
0.1076 |
11.47 |
11600 |
0.4298 |
0.2686 |
0.1124 |
11.87 |
12000 |
0.4025 |
0.2704 |
0.1043 |
12.26 |
12400 |
0.4209 |
0.2659 |
0.0976 |
12.66 |
12800 |
0.4070 |
0.2672 |
0.1012 |
13.06 |
13200 |
0.4161 |
0.2720 |
0.0872 |
13.45 |
13600 |
0.4245 |
0.2697 |
0.0933 |
13.85 |
14000 |
0.4295 |
0.2684 |
0.0881 |
14.24 |
14400 |
0.4011 |
0.2650 |
0.0848 |
14.64 |
14800 |
0.3991 |
0.2675 |
0.0852 |
15.03 |
15200 |
0.4166 |
0.2617 |
0.0825 |
15.43 |
15600 |
0.4188 |
0.2639 |
0.081 |
15.83 |
16000 |
0.4181 |
0.2547 |
0.0753 |
16.22 |
16400 |
0.4103 |
0.2560 |
0.0747 |
16.62 |
16800 |
0.4017 |
0.2498 |
0.0761 |
17.01 |
17200 |
0.4159 |
0.2563 |
0.0711 |
17.41 |
17600 |
0.4112 |
0.2603 |
0.0698 |
17.8 |
18000 |
0.4335 |
0.2529 |
0.073 |
18.2 |
18400 |
0.4120 |
0.2512 |
0.0665 |
18.6 |
18800 |
0.4335 |
0.2496 |
0.0657 |
18.99 |
19200 |
0.4143 |
0.2468 |
0.0617 |
19.39 |
19600 |
0.4339 |
0.2435 |
0.06 |
19.78 |
20000 |
0.4179 |
0.2438 |
0.0613 |
20.18 |
20400 |
0.4251 |
0.2393 |
0.0583 |
20.57 |
20800 |
0.4347 |
0.2422 |
0.0562 |
20.97 |
21200 |
0.4246 |
0.2377 |
0.053 |
21.36 |
21600 |
0.4198 |
0.2338 |
0.0525 |
21.76 |
22000 |
0.4511 |
0.2427 |
0.0499 |
22.16 |
22400 |
0.4482 |
0.2353 |
0.0475 |
22.55 |
22800 |
0.4449 |
0.2329 |
0.0465 |
22.95 |
23200 |
0.4364 |
0.2320 |
0.0443 |
23.34 |
23600 |
0.4481 |
0.2304 |
0.0458 |
23.74 |
24000 |
0.4442 |
0.2267 |
0.0453 |
24.13 |
24400 |
0.4402 |
0.2261 |
0.0426 |
24.53 |
24800 |
0.4262 |
0.2232 |
0.0431 |
24.93 |
25200 |
0.4251 |
0.2210 |
0.0389 |
25.32 |
25600 |
0.4455 |
0.2232 |
0.039 |
25.72 |
26000 |
0.4372 |
0.2236 |
0.0378 |
26.11 |
26400 |
0.4236 |
0.2212 |
0.0348 |
26.51 |
26800 |
0.4359 |
0.2204 |
0.0361 |
26.9 |
27200 |
0.4248 |
0.2192 |
0.0356 |
27.3 |
27600 |
0.4397 |
0.2184 |
0.0325 |
27.7 |
28000 |
0.4367 |
0.2181 |
0.0313 |
28.09 |
28400 |
0.4477 |
0.2136 |
0.0306 |
28.49 |
28800 |
0.4533 |
0.2135 |
0.0314 |
28.88 |
29200 |
0.4410 |
0.2136 |
0.0307 |
29.28 |
29600 |
0.4457 |
0.2113 |
0.0309 |
29.67 |
30000 |
0.4426 |
0.2117 |
Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.17.1.dev0
- Tokenizers 0.11.0
đ License
This project uses the Apache 2.0 license.