speecht5_finetuned_kazakh_tts2_1

This model is a fine-tuned version of microsoft/speecht5_tts on the KazakhTTS2 dataset (Mussakhojayeva, S., Khassanov, Y., & Varol, H.A. (2022). KazakhTTS2: Extending the Open-Source Kazakh TTS Corpus With More Data, Speakers, and Topics. International Conference on Language Resources and Evaluation). It achieves the following results on the evaluation set:

• Loss: 0.4600

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: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.725	0.06	100	0.6639
0.6132	0.11	200	0.5466
0.571	0.17	300	0.5207
0.5647	0.22	400	0.5120
0.5556	0.28	500	0.5047
0.5475	0.34	600	0.5003
0.5432	0.39	700	0.4975
0.5366	0.45	800	0.4944
0.5376	0.5	900	0.4913
0.5325	0.56	1000	0.4868
0.5281	0.62	1100	0.4861
0.5288	0.67	1200	0.4848
0.5251	0.73	1300	0.4825
0.5213	0.78	1400	0.4818
0.5225	0.84	1500	0.4823
0.5199	0.9	1600	0.4812
0.5211	0.95	1700	0.4816
0.5194	1.01	1800	0.4826
0.5224	1.06	1900	0.4798
0.5213	1.12	2000	0.4800

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1+cu118
• Datasets 2.18.0
• Tokenizers 0.15.2