Whisper Small Catalan
This is an automatic speech recognition model that also does punctuation and casing. This model is for research only, we do not recommend using this model on production environments. See our learnings when training these models.
This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 ca dataset. It achieves the following results on the evaluation set:
- Loss: 0.1980
- Wer: 8.5695
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: 20000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
0.2128 | 0.1 | 2000 | 0.2644 | 13.0303 |
0.1361 | 1.1 | 4000 | 0.2300 | 10.9568 |
0.0658 | 2.1 | 6000 | 0.2376 | 11.2810 |
0.102 | 3.09 | 8000 | 0.2156 | 9.8730 |
0.0706 | 4.09 | 10000 | 0.2126 | 9.6179 |
0.0428 | 5.09 | 12000 | 0.2178 | 9.3405 |
0.0503 | 6.09 | 14000 | 0.2109 | 9.1356 |
0.0778 | 7.08 | 16000 | 0.2058 | 9.2001 |
0.0082 | 8.08 | 18000 | 0.2173 | 8.9941 |
0.0994 | 9.08 | 20000 | 0.1980 | 8.5695 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.10.0+cu102
- Datasets 2.7.1
- Tokenizers 0.13.2
- Downloads last month
- 12
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Dataset used to train softcatala/whisper-small-ca
Evaluation results
- Wer on mozilla-foundation/common_voice_11_0 caself-reported8.569
- Wer on google/fleurs caself-reported10.640
- Wer on projecte-aina/parlament_parla cleanself-reported19.000