wav2vec2-large-xls-r-300m-telugu-asr

This model is a fine-tuned version of henilp105/wav2vec2-large-xls-r-300m-telugu-asr on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.1050
• Wer: 0.6656

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.0506	2.3	200	0.8841	0.7564
0.6354	4.59	400	0.7448	0.6912
0.3934	6.89	600	0.8321	0.6929
0.2652	9.19	800	0.9529	0.6984
0.2022	11.49	1000	0.9490	0.6979
0.1514	13.79	1200	1.0025	0.6869
0.124	16.09	1400	1.0367	0.6799
0.1007	18.39	1600	1.0658	0.6734
0.0875	20.69	1800	1.0758	0.6779
0.0838	22.98	2000	1.0999	0.6701
0.0745	25.29	2200	1.1020	0.6708
0.0641	27.58	2400	1.1140	0.6683
0.0607	29.88	2600	1.1050	0.6656

Framework versions

• Transformers 4.24.0
• Pytorch 1.10.0+cu113
• Datasets 1.18.3
• Tokenizers 0.13.2