Arbi-Houssem/speecht5_ar_tn_1.5

SpeechT5 TTS Tunisien

This model is a fine-tuned version of MBZUAI/speecht5_tts_clartts_ar on the Tunisian_dataset_STT-TTS15s_filtred1.0_Mixed dataset. It achieves the following results on the evaluation set:

• Loss: 0.4403

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.6355	6.6667	500	0.5657
0.5548	13.3333	1000	0.5314
0.539	20.0	1500	0.5286
0.5475	26.6667	2000	0.5220
0.5031	33.3333	2500	0.5193
0.5077	40.0	3000	0.5174
0.4724	46.6667	3500	0.4577
0.54	53.3333	4000	0.4610
0.5289	60.0	4500	0.4431
0.53	66.6667	5000	0.4414
0.5029	73.3333	5500	0.4404
0.4944	80.0	6000	0.4406
0.508	86.6667	6500	0.4385
0.5019	93.3333	7000	0.4410
0.4639	100.0	7500	0.4394
0.4918	106.6667	8000	0.4408
0.4914	113.3333	8500	0.4401
0.4747	120.0	9000	0.4413
0.4886	126.6667	9500	0.4409
0.4504	133.3333	10000	0.4403

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1