Uzbek Stt

🚀 Uzbek Speech-to-Text Model

This is the best version of the Speech-to-Text (STT) developed by the Oyqiz team! It offers high - quality automatic speech recognition capabilities, trained on a specific dataset to achieve excellent performance.

🚀 Quick Start

This model is trained for 100 epochs using facebook/wav2vec2-xls-r-300m and the MOZILLA - FOUNDATION/COMMON_VOICE_10_0 - UZ dataset.

Training Results

• Error: 0.1963
• Word Error Rate: 0.2102

Training Hyperparameters

The hyperparameters used for training are as follows:

• 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

✨ Features

• High - Quality Training: Trained on a large - scale dataset for 100 epochs to ensure high accuracy.
• Advanced Model: Based on the facebook/wav2vec2-xls-r-300m architecture.

📄 License

This project is licensed under the Apache - 2.0 license.

👨‍💻 Team Members

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