|
|
--- |
|
|
base_model: openai/whisper-medium |
|
|
datasets: |
|
|
- generator |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- generated_from_trainer |
|
|
model-index: |
|
|
- name: whisper-medium-sb-lug-eng |
|
|
results: [] |
|
|
--- |
|
|
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
|
|
# whisper-medium-sb-lug-eng |
|
|
|
|
|
This model is a fine-tuned version of [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) on the generator dataset. |
|
|
It achieves the following results on the evaluation set: |
|
|
- Loss: 0.1181 |
|
|
- Wer Lug: 0.458 |
|
|
- Wer Eng: 0.023 |
|
|
- Wer Mean: 0.24 |
|
|
- Cer Lug: 0.273 |
|
|
- Cer Eng: 0.011 |
|
|
- Cer Mean: 0.142 |
|
|
|
|
|
## Model description |
|
|
|
|
|
More information needed |
|
|
|
|
|
## Intended uses & limitations |
|
|
|
|
|
More information needed |
|
|
|
|
|
## Training and evaluation data |
|
|
|
|
|
More information needed |
|
|
|
|
|
## Training procedure |
|
|
|
|
|
### Training hyperparameters |
|
|
|
|
|
The following hyperparameters were used during training: |
|
|
- learning_rate: 1e-05 |
|
|
- train_batch_size: 16 |
|
|
- eval_batch_size: 8 |
|
|
- seed: 42 |
|
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
|
- lr_scheduler_type: linear |
|
|
- lr_scheduler_warmup_steps: 500 |
|
|
- training_steps: 12000 |
|
|
- mixed_precision_training: Native AMP |
|
|
|
|
|
### Training results |
|
|
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Wer Lug | Wer Eng | Wer Mean | Cer Lug | Cer Eng | Cer Mean | |
|
|
|:-------------:|:------:|:-----:|:---------------:|:-------:|:-------:|:--------:|:-------:|:-------:|:--------:| |
|
|
| 0.6895 | 0.0417 | 500 | 0.2672 | 0.398 | 0.031 | 0.214 | 0.084 | 0.016 | 0.05 | |
|
|
| 0.5359 | 0.0833 | 1000 | 0.1889 | 0.594 | 0.027 | 0.311 | 0.287 | 0.012 | 0.149 | |
|
|
| 0.4759 | 0.125 | 1500 | 0.1722 | 0.216 | 0.027 | 0.121 | 0.12 | 0.011 | 0.066 | |
|
|
| 0.4283 | 0.1667 | 2000 | 0.1595 | 0.224 | 0.024 | 0.124 | 0.062 | 0.012 | 0.037 | |
|
|
| 0.3989 | 0.2083 | 2500 | 0.1502 | 0.193 | 0.026 | 0.109 | 0.062 | 0.011 | 0.037 | |
|
|
| 0.3676 | 0.25 | 3000 | 0.1501 | 0.166 | 0.028 | 0.097 | 0.049 | 0.013 | 0.031 | |
|
|
| 0.3475 | 0.2917 | 3500 | 0.1424 | 0.184 | 0.028 | 0.106 | 0.066 | 0.013 | 0.04 | |
|
|
| 0.3469 | 0.3333 | 4000 | 0.1388 | 0.246 | 0.029 | 0.138 | 0.111 | 0.013 | 0.062 | |
|
|
| 0.3115 | 0.375 | 4500 | 0.1355 | 0.916 | 0.029 | 0.473 | 0.541 | 0.013 | 0.277 | |
|
|
| 0.2969 | 0.4167 | 5000 | 0.1343 | 0.304 | 0.028 | 0.166 | 0.137 | 0.009 | 0.073 | |
|
|
| 0.2438 | 1.0383 | 5500 | 0.1246 | 0.191 | 0.027 | 0.109 | 0.081 | 0.011 | 0.046 | |
|
|
| 0.237 | 1.08 | 6000 | 0.1293 | 0.193 | 0.026 | 0.109 | 0.081 | 0.01 | 0.045 | |
|
|
| 0.2192 | 1.1217 | 6500 | 0.1314 | 0.199 | 0.022 | 0.111 | 0.09 | 0.008 | 0.049 | |
|
|
| 0.2404 | 1.1633 | 7000 | 0.1308 | 0.136 | 0.028 | 0.082 | 0.049 | 0.01 | 0.03 | |
|
|
| 0.2185 | 1.205 | 7500 | 0.1283 | 0.179 | 0.019 | 0.099 | 0.068 | 0.007 | 0.038 | |
|
|
| 0.22 | 1.2467 | 8000 | 0.1235 | 0.272 | 0.025 | 0.149 | 0.164 | 0.01 | 0.087 | |
|
|
| 0.2252 | 1.2883 | 8500 | 0.1263 | 0.243 | 0.023 | 0.133 | 0.108 | 0.009 | 0.059 | |
|
|
| 0.2121 | 1.33 | 9000 | 0.1238 | 0.549 | 0.037 | 0.293 | 0.27 | 0.019 | 0.145 | |
|
|
| 0.2106 | 1.3717 | 9500 | 0.1213 | 0.23 | 0.024 | 0.127 | 0.112 | 0.009 | 0.06 | |
|
|
| 0.2095 | 1.4133 | 10000 | 0.1201 | 0.305 | 0.027 | 0.166 | 0.169 | 0.013 | 0.091 | |
|
|
| 0.1517 | 2.035 | 10500 | 0.1206 | 0.376 | 0.026 | 0.201 | 0.212 | 0.011 | 0.111 | |
|
|
| 0.1639 | 2.0767 | 11000 | 0.1200 | 0.415 | 0.022 | 0.219 | 0.262 | 0.01 | 0.136 | |
|
|
| 0.1561 | 2.1183 | 11500 | 0.1185 | 0.497 | 0.023 | 0.26 | 0.301 | 0.011 | 0.156 | |
|
|
| 0.1634 | 2.16 | 12000 | 0.1181 | 0.458 | 0.023 | 0.24 | 0.273 | 0.011 | 0.142 | |
|
|
|
|
|
|
|
|
### Framework versions |
|
|
|
|
|
- Transformers 4.42.3 |
|
|
- Pytorch 2.2.0 |
|
|
- Datasets 2.20.0 |
|
|
- Tokenizers 0.19.1 |
|
|
|
