|
--- |
|
license: apache-2.0 |
|
base_model: openai/whisper-medium |
|
tags: |
|
- generated_from_trainer |
|
datasets: |
|
- generator |
|
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: 0.1179 |
|
- Wer Lug: 0.558 |
|
- Wer Eng: 0.025 |
|
- Wer Mean: 0.292 |
|
- Cer Lug: 0.285 |
|
- Cer Eng: 0.011 |
|
- Cer Mean: 0.148 |
|
|
|
## 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 | |
|
|:-------------:|:------:|:-----:|:---------------:|:-------:|:-------:|:--------:|:-------:|:-------:|:--------:| |
|
| 0.7169 | 0.0417 | 500 | 0.2614 | 0.715 | 0.039 | 0.377 | 0.264 | 0.02 | 0.142 | |
|
| 0.5343 | 0.0833 | 1000 | 0.1939 | 0.391 | 0.029 | 0.21 | 0.147 | 0.013 | 0.08 | |
|
| 0.4529 | 0.125 | 1500 | 0.1864 | 0.256 | 0.032 | 0.144 | 0.069 | 0.015 | 0.042 | |
|
| 0.3733 | 0.1667 | 2000 | 0.1597 | 0.276 | 0.024 | 0.15 | 0.093 | 0.01 | 0.052 | |
|
| 0.3874 | 0.2083 | 2500 | 0.1520 | 0.733 | 0.023 | 0.378 | 0.314 | 0.01 | 0.162 | |
|
| 0.3641 | 0.25 | 3000 | 0.1408 | 0.762 | 0.027 | 0.395 | 0.339 | 0.012 | 0.176 | |
|
| 0.3682 | 0.2917 | 3500 | 0.1413 | 0.956 | 0.016 | 0.486 | 0.503 | 0.006 | 0.254 | |
|
| 0.3379 | 0.3333 | 4000 | 0.1376 | 0.358 | 0.027 | 0.192 | 0.177 | 0.01 | 0.094 | |
|
| 0.3324 | 0.375 | 4500 | 0.1317 | 0.249 | 0.022 | 0.135 | 0.107 | 0.007 | 0.057 | |
|
| 0.3003 | 0.4167 | 5000 | 0.1304 | 0.403 | 0.025 | 0.214 | 0.174 | 0.009 | 0.091 | |
|
| 0.2259 | 1.0383 | 5500 | 0.1294 | 0.608 | 0.021 | 0.314 | 0.263 | 0.007 | 0.135 | |
|
| 0.2259 | 1.08 | 6000 | 0.1309 | 0.835 | 0.025 | 0.43 | 0.417 | 0.008 | 0.213 | |
|
| 0.2531 | 1.1217 | 6500 | 0.1252 | 0.249 | 0.022 | 0.135 | 0.109 | 0.008 | 0.058 | |
|
| 0.2258 | 1.1633 | 7000 | 0.1259 | 0.384 | 0.023 | 0.203 | 0.178 | 0.007 | 0.093 | |
|
| 0.2205 | 1.205 | 7500 | 0.1264 | 0.376 | 0.021 | 0.198 | 0.181 | 0.007 | 0.094 | |
|
| 0.2235 | 1.2467 | 8000 | 0.1274 | 0.395 | 0.025 | 0.21 | 0.196 | 0.008 | 0.102 | |
|
| 0.2219 | 1.2883 | 8500 | 0.1252 | 0.349 | 0.029 | 0.189 | 0.162 | 0.01 | 0.086 | |
|
| 0.2334 | 1.33 | 9000 | 0.1224 | 0.292 | 0.028 | 0.16 | 0.128 | 0.01 | 0.069 | |
|
| 0.2339 | 1.3717 | 9500 | 0.1207 | 0.305 | 0.026 | 0.166 | 0.136 | 0.01 | 0.073 | |
|
| 0.2094 | 1.4133 | 10000 | 0.1194 | 0.569 | 0.023 | 0.296 | 0.276 | 0.009 | 0.143 | |
|
| 0.1643 | 2.035 | 10500 | 0.1184 | 0.818 | 0.026 | 0.422 | 0.466 | 0.011 | 0.239 | |
|
| 0.1549 | 2.0767 | 11000 | 0.1196 | 0.651 | 0.022 | 0.336 | 0.374 | 0.008 | 0.191 | |
|
| 0.1727 | 2.1183 | 11500 | 0.1182 | 0.547 | 0.025 | 0.286 | 0.29 | 0.011 | 0.15 | |
|
| 0.1617 | 2.16 | 12000 | 0.1179 | 0.558 | 0.025 | 0.292 | 0.285 | 0.011 | 0.148 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.42.3 |
|
- Pytorch 2.2.0 |
|
- Datasets 2.20.0 |
|
- Tokenizers 0.19.1 |
|
|
