oyqiz/uzbek_stt_4_version

O'zbekcha Speech To Text 4-versiyasi

Bu modelimiz facebook/wav2vec2-base va MOZILLA-FOUNDATION/COMMON_VOICE_10_0 - ning O'zbekcha dataseti bilan o'qitildi. Model 100ta epoxda o'qitildi. O'qitish jarayonida quyidagi natijalarga erishildi:

• Xatolik: 0.7755
• So'zning xatolik darajasi: 0.3976

O'qitish uchun belgilangan giperparameterlar

Quyidagi giperparameterlar o'qitish uchun ishlatildi:

• learning_rate: 3e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 16
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100.0
• mixed_precision_training: Native AMP

O'qitish natijalari

Xatolik	Epox	Qadam	Tasdiq xatoligi	SXD
0.1532	13.59	20000	0.6070	0.4822
0.099	27.17	40000	0.6438	0.4461
0.0769	40.76	60000	0.6889	0.4343
0.0622	54.35	80000	0.7638	0.4181
0.0561	67.93	100000	0.7523	0.4081
0.0454	81.52	120000	0.7569	0.4006
0.0378	95.11	140000	0.7702	0.3989