augustinjianu/whisper-tiny-ro

Whisper Tiny Ro (local) - Augustin Jianu

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5978
• Wer: 37.4835

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: 8
• seed: 42
• optimizer: Use OptimizerNames.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
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4417	1.7730	1000	0.5327	43.8513
0.1813	3.5461	2000	0.4666	38.8689
0.0751	5.3191	3000	0.4645	36.5006
0.0326	7.0922	4000	0.4803	36.4614
0.0234	8.8652	5000	0.5087	36.5148
0.0082	10.6383	6000	0.5424	36.6252
0.0042	12.4113	7000	0.5650	37.6509
0.0029	14.1844	8000	0.5809	36.8710
0.0025	15.9574	9000	0.5922	38.1495
0.0021	17.7305	10000	0.5978	37.4835

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0