Oyqiz jamoasi a'zolari tomonidan qilingan STT ning eng yaxshi versiyasi!

Foziljon To'lqinov, Shaxboz Zohidov, Abduraxim Jabborov, Yahyoxon Rahimov, Mahmud Jumanazarov

Bu model facebook/wav2vec2-xls-r-300m va MOZILLA-FOUNDATION/COMMON_VOICE_10_0 - UZ dataseti bilan 100ta epoxta o'qitilgan. O'qitish natijalari:

• Xatolik: 0.1963
• So'z xatoligi: 0.2102

O'qitish giperparameterlari

O'qitish uchun ishlatilgan giperparameterlar:

• learning_rate: 0.00003
• 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