ptah23/speecht5_finetuned_voxpopuli_es

speec T5 ES - Peter Gelderbloem

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

• Loss: 0.4400

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: 4
• eval_batch_size: 4
• 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
0.6709	0.14	100	0.6107
0.624	0.28	200	0.5773
0.5921	0.43	300	0.5397
0.5504	0.57	400	0.4941
0.5289	0.71	500	0.4807
0.5236	0.85	600	0.4732
0.5194	0.99	700	0.4672
0.5095	1.13	800	0.4650
0.508	1.28	900	0.4608
0.5029	1.42	1000	0.4580
0.5029	1.56	1100	0.4569
0.4952	1.7	1200	0.4540
0.4939	1.84	1300	0.4540
0.4924	1.99	1400	0.4523
0.4902	2.13	1500	0.4519
0.492	2.27	1600	0.4491
0.4896	2.41	1700	0.4497
0.4874	2.55	1800	0.4482
0.4889	2.7	1900	0.4473
0.4888	2.84	2000	0.4471
0.4902	2.98	2100	0.4461
0.4818	3.12	2200	0.4457
0.4854	3.26	2300	0.4451
0.4886	3.4	2400	0.4429
0.4799	3.55	2500	0.4429
0.4806	3.69	2600	0.4429
0.4777	3.83	2700	0.4416
0.4787	3.98	2800	0.4423
0.4823	4.12	2900	0.4419
0.4822	4.26	3000	0.4409
0.4796	4.4	3100	0.4404
0.4805	4.54	3200	0.4418
0.4795	4.69	3300	0.4409
0.4784	4.83	3400	0.4411
0.4812	4.97	3500	0.4406
0.478	5.11	3600	0.4399
0.4766	5.25	3700	0.4396
0.4821	5.39	3800	0.4407
0.48	5.54	3900	0.4396
0.4795	5.68	4000	0.4400

Framework versions

• Transformers 4.30.2
• Pytorch 1.12.1+cu116
• Datasets 2.4.0
• Tokenizers 0.12.1