whisper-base-cs

This model is a fine-tuned version of openai/whisper-base on the CommonVoice11 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3785
• Wer: 35.4791

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: 4000
• mixed_precision_training: Native AMP

Training results

BASE-CS baseline performance: {'eval_loss': 2.514087438583374, 'eval_wer': 82.45662504144104, 'eval_runtime': 2620.0407, 'eval_samples_per_second': 2.944, 'eval_steps_per_second': 0.368}

Training Loss	Epoch	Step	Validation Loss	Wer
------	0.00	0000	2.5141	82.4566
0.2991	1.44	1000	0.4414	42.2993
0.1776	2.89	2000	0.3818	36.7573
0.0916	4.33	3000	0.3774	35.6080
0.076	5.78	4000	0.3785	35.4791

Framework versions

• Transformers 4.36.0
• Pytorch 2.1.0+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0