wav2vec2-large-xlsr-53-Bengali

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4707
• Wer: 0.5156

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: 0.0003
• train_batch_size: 24
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 48
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.5484	2.85	500	1.6415	1.0493
0.8446	5.71	1000	0.5394	0.6886
0.4604	8.57	1500	0.4741	0.6090
0.3466	11.43	2000	0.4685	0.5788
0.2707	14.28	2500	0.4741	0.5449
0.2317	17.14	3000	0.4661	0.5157
0.2024	20.0	3500	0.4707	0.5156

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.1+cu116
• Datasets 2.11.0
• Tokenizers 0.13.2