speecht5_finetuned_voxpopuli
This model is a fine-tuned version of microsoft/speecht5_tts on the voxpopuli dataset. It achieves the following results on the evaluation set:
- validation Loss: 0.5676
- training loss: 0.38
Model description
text-to-speech
Intended uses & limitations
text to speech, stst models
Training and evaluation data
finetuning using the voxpopuli dataset for the Lithuanian language, in this case there were few speakers and few examples, so the training gives us 0.56 validation loss and 0.38 of training loss, This means the model may not generalize well to new data it hasn't seen before. To avoid overfitting, you can try some regularization techniques, such as dropout, batch normalization, or model size reduction.
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 4
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.443 | 380.95 | 1000 | 0.5600 |
| 0.4045 | 761.9 | 2000 | 0.5717 |
| 0.3877 | 1142.86 | 3000 | 0.5647 |
| 0.3845 | 1523.81 | 4000 | 0.5676 |
Framework versions
- Transformers 4.30.2
- Pytorch 2.0.0
- Datasets 2.1.0
- Tokenizers 0.13.3