whisper-base-cs-cv11-timestetch02-gain01-pitch02-gaussian02-lowpass01-timemask020-freqmask020

This model is a fine-tuned version of openai/whisper-base on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3065
• Wer: 31.6536

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8775	0.72	1000	0.4722	45.7067
0.6983	1.44	2000	0.3854	38.8275
0.5251	2.17	3000	0.3505	35.8143
0.5001	2.89	4000	0.3291	33.8619
0.4956	3.61	5000	0.3192	33.0405
0.4256	4.33	6000	0.3110	32.4364
0.398	5.06	7000	0.3081	32.0588
0.3921	5.78	8000	0.3065	31.6536

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.0+cu121
• Datasets 2.17.1
• Tokenizers 0.15.2