wav2vec2-large-xlsr-53-Bangla

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: 0.6125
• Wer: 0.5442

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: 0.0004
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.6881	2.28	600	1.0325	0.9634
0.8087	4.56	1200	0.6090	0.7430
0.5089	6.84	1800	0.5156	0.6615
0.3864	9.13	2400	0.5287	0.6676
0.3064	11.41	3000	0.5411	0.6278
0.2535	13.69	3600	0.5206	0.6149
0.216	15.97	4200	0.5596	0.6120
0.1852	18.25	4800	0.5658	0.5821
0.1653	20.53	5400	0.5938	0.5521
0.1499	22.81	6000	0.5825	0.5645
0.1323	25.09	6600	0.6151	0.5593
0.122	27.38	7200	0.6046	0.5556
0.1118	29.66	7800	0.6125	0.5442

Framework versions

• Transformers 4.24.0
• Pytorch 2.0.0+cu118
• Datasets 2.11.0
• Tokenizers 0.13.3