This model is a fine-tuned version of cahya/wav2vec2-base-turkish-artificial-cv on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:
| Dataset | WER | CER | |
|---|---|---|---|
| 1 | Common Voice 6.1 | 9.437 | 3.325 |
| 2 | Common Voice 7.0 | 8.147 | 2.802 |
| 3 | Common Voice 8.0 | 8.335 | 2.336 |
| 4 | Speech Recognition Community | 28.011 | 10.66 |
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
The following datasets were used for finetuning:
- Common Voice 7.0 TR 'train', 'validation' and 'other' split were used for training.
- Media Speech
- Magic Hub
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 7.5e-06
- train_batch_size: 6
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 24
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 5.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 1.1224 | 3.45 | 500 | 0.1641 | 0.1396 |
Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.18.2
- Tokenizers 0.10.3
- Downloads last month
- 48
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 cahya/wav2vec2-base-turkish
Evaluation results
- Test WER on Common Voice 6.1self-reported9.437
- Test CER on Common Voice 6.1self-reported3.325
- Test WER on Common Voice 7self-reported8.147
- Test CER on Common Voice 7self-reported2.802
- Test WER on Robust Speech Event - Dev Dataself-reported28.011
- Test CER on Robust Speech Event - Dev Dataself-reported10.660
- Test WER on Robust Speech Event - Test Dataself-reported33.620