wav2vec2-xls-r-300m-crh

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the custom Crimean Tatar dataset. It achieves the following results on the evaluation set:

• Loss: 0.738475
• Wer: 0.4494
• Cer: 0.1254

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: 3e-05
• train_batch_size: 24
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 6
• total_train_batch_size: 144
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.0+cu117
• Datasets 2.6.1
• Tokenizers 0.13.1