bangla_voice

This model is a fine-tuned version of iftekher/bangla_voice on the None dataset. It achieves the following results on the evaluation set:

• Loss: 208.2614
• Wer: 0.3201

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: 0.0001
• train_batch_size: 16
• 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: 100
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
158.927	0.21	100	81.4025	0.3489
206.3938	0.42	200	117.4497	0.3680
194.8868	0.64	300	473.2094	0.3622
177.3037	0.85	400	81.0834	0.3585
150.9285	1.06	500	397.6080	0.3592
164.899	1.27	600	71.5732	0.3476
157.9872	1.48	700	76.6225	0.3560
139.5956	1.69	800	76.4330	0.3512
132.7378	1.91	900	154.8127	0.3378
137.2875	2.12	1000	275.6554	0.3453
128.1135	2.33	1100	210.1160	0.3409
124.5749	2.54	1200	109.8560	0.3400
115.9728	2.75	1300	165.5507	0.3373
120.9464	2.97	1400	248.8096	0.3357
104.8963	3.18	1500	308.7221	0.3361
115.9144	3.39	1600	214.0615	0.3300
109.0966	3.6	1700	197.1803	0.3286
111.4354	3.81	1800	189.1278	0.3245
111.9318	4.03	1900	191.4921	0.3282
109.2148	4.24	2000	185.1797	0.3298
114.0561	4.45	2100	190.5829	0.3229
105.7045	4.66	2200	209.0799	0.3220
127.4207	4.87	2300	208.2614	0.3201

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.13.3
• Tokenizers 0.10.3