wav2vec2-base-Telugu-large

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3811
• Wer: 0.3630
• Cer: 0.0716

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: 6
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
7.205	1.9737	300	3.2285	1.0	1.0
1.3015	3.9474	600	0.7145	0.7256	0.1790
0.6733	5.9211	900	0.5099	0.6079	0.1339
0.5037	7.8947	1200	0.4510	0.5383	0.1146
0.4039	9.8684	1500	0.3932	0.4957	0.1011
0.3224	11.8421	1800	0.3733	0.4537	0.0909
0.2699	13.8158	2100	0.3685	0.4291	0.0840
0.2177	15.7895	2400	0.3664	0.3836	0.0747
0.19	17.7632	2700	0.3722	0.3821	0.0743
0.1596	19.7368	3000	0.3811	0.3630	0.0716

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 1.18.3
• Tokenizers 0.19.1