This model was trained using the instructions provided on this notebook but using the catalan subset of OpenSLR dataset. The main change is the use of trimming to delete large parts of silence that this dataset originally have. You can check the notebook used for this training here

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
gradient_accumulation_steps: 2
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: 8000

Training results

Training Loss	Epoch	Step	Validation Loss
0.5039	8.37	1000	0.4530
0.4723	16.74	2000	0.4345
0.4583	25.1	3000	0.4316
0.4565	33.47	4000	0.4294
0.4363	41.84	5000	0.4329
0.446	50.21	6000	0.4331
0.4508	58.58	7000	0.4336
0.4529	66.95	8000	0.4360

Framework versions

Transformers 4.32.0.dev0
Pytorch 2.0.1+cu118
Datasets 2.14.4
Tokenizers 0.13.3

wetdog/speecht5_tts_openslr_ca

SpeechT5 TTS Catalan

Model description