liepa3-whisper-base-lt-liepa2_191_minimized_v11

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

• Loss: 0.4414
• Wer: 41.1565

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: 16
• eval_batch_size: 1
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3585	0.0549	500	0.9896	74.9860
0.7078	0.1097	1000	0.7824	64.8662
0.589	0.1646	1500	0.6871	58.4774
0.5194	0.2195	2000	0.6319	54.9628
0.4762	0.2743	2500	0.5909	51.3903
0.4462	0.3292	3000	0.5629	50.4844
0.4228	0.3841	3500	0.5404	48.8600
0.4073	0.4389	4000	0.5208	46.8132
0.3864	0.4938	4500	0.5044	45.6792
0.3709	0.5487	5000	0.4914	44.4736
0.3591	0.6035	5500	0.4806	44.0343
0.3558	0.6584	6000	0.4730	43.0858
0.3421	0.7133	6500	0.4642	42.6584
0.3403	0.7681	7000	0.4580	42.6567
0.3311	0.8230	7500	0.4508	42.0607
0.3303	0.8779	8000	0.4471	41.6027
0.3251	0.9327	8500	0.4434	41.1174
0.325	0.9876	9000	0.4414	41.1565

Framework versions

• Transformers 4.51.3
• Pytorch 2.2.1
• Datasets 3.6.0
• Tokenizers 0.21.2