Whisper Small Swedish
This model is an adapted version of openai/whisper-small on the Common Voice 11.0 dataset in Swedish. It achieves the following results on the evaluation set:
- Wer: 19.8166
Model description & uses
This model is the openai whisper small transformer adapted for Swedish audio to text transcription. The model is available through its HuggingFace web app
Training and evaluation data
Data used for training is the initial 10% of train and validation of Swedish Common Voice 11.0 from Mozilla Foundation. The dataset used for evaluation is the initial 10% of test of Swedish Common Voice. The training data has been augmented with random noise, random pitching and change of the speed of the voice.
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 8
- eval_batch_size: 8
- lr_scheduler_type: constant
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
- weight decay: 0
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1379 | 0.95 | 1000 | 0.295811 | 21.467 |
| 0.0245 | 2.86 | 3000 | 0.300059 | 20.160 |
| 0.0060 | 3.82 | 4000 | 0.320301 | 19.762 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2
