Whisper-med-ml
This model is a fine-tuned version of openai/whisper-medium on the datasets: IMASC, MSC, OpenSLR Malayalam Train split, Festvox Malayalam .
It achieves the following results on the validation set : OpenSLR-Test
- Loss: 0.0318
- Wer: 14.7300
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: 1000
- training_steps: 6000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
0.0599 | 0.4 | 1000 | 0.0910 | 42.4981 |
0.0341 | 0.79 | 2000 | 0.0584 | 30.0572 |
0.0183 | 1.19 | 3000 | 0.0439 | 23.1650 |
0.0147 | 1.58 | 4000 | 0.0363 | 18.7360 |
0.0107 | 1.98 | 5000 | 0.0322 | 16.4220 |
0.0032 | 2.37 | 6000 | 0.0318 | 14.7300 |
Framework versions
- Transformers 4.37.2
- Pytorch 2.1.1+cu121
- Datasets 2.16.1
- Tokenizers 0.15.1
- Downloads last month
- 17
Finetuned from
Datasets used to train vrclc/Whisper-med-ml
Space using vrclc/Whisper-med-ml 1
Evaluation results
- WER on Common Voice 13 Malayalamtest set self-reported63.640
- CER on Common Voice 13 Malayalamtest set self-reported13.610
- WER on Common Voice 16 Malayalamtest set self-reported64.630
- CER on Common Voice 16 Malayalamtest set self-reported14.070
- WER on OpenSLR Malayalam -Testtest set self-reported14.650
- CER on OpenSLR Malayalam -Testtest set self-reported2.590