ccibeekeoc42/speecht5_finetuned_naija_ig_yo_2025-01-20

SpeechT5 TTS Igb0 Yoruba

This model is a fine-tuned version of microsoft/speecht5_tts on the naija_tts_concatenated dataset. It achieves the following results on the evaluation set:

• Loss: 0.4370

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: 6
• eval_batch_size: 6
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 12
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.5354	0.1339	500	0.4752
0.5157	0.2679	1000	0.4604
0.5028	0.4018	1500	0.4537
0.4872	0.5358	2000	0.4489
0.4892	0.6697	2500	0.4453
0.4813	0.8036	3000	0.4436
0.483	0.9376	3500	0.4421
0.4763	1.0715	4000	0.4401
0.4786	1.2055	4500	0.4400
0.481	1.3394	5000	0.4379
0.4811	1.4733	5500	0.4393
0.4846	1.6073	6000	0.4370

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0