whisper-tiny-fluers_V2_telugu_Augmentation_full_datset_V2_e5

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3374
• Wer: 61.2000

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: 1.5e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-06
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2926	0.09	300	1.3993	129.5
1.0948	0.18	600	1.2674	109.1500
0.6591	0.28	900	0.5519	81.55
0.5326	0.37	1200	0.4361	72.55
0.4737	0.46	1500	0.4036	69.2000
0.4239	0.55	1800	0.3793	63.6
0.4011	0.64	2100	0.3625	62.2500
0.3687	0.73	2400	0.3651	62.5
0.3712	0.83	2700	0.3491	59.9
0.3686	0.92	3000	0.3438	60.6500
0.3381	1.01	3300	0.3391	58.25
0.3483	1.1	3600	0.3385	60.0500
0.341	1.19	3900	0.3374	61.2000

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 1.12.1
• Datasets 2.11.0
• Tokenizers 0.13.3