arielcerdap/whisper-large-v3-turbo-stutter-optimized

Whisper Large V3 Turbo Optimized for Stuttered Speech

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the TimeStamped dataset. It achieves the following results on the evaluation set:

• Loss: 1.9310
• Wer: 12.9363
• Wer Ortho: 7.2442
• Cer: 7.2141

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: 8e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 8000
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho	Cer
1.5069	5.8187	500	1.6693	15.0754	7.2766	6.8200
1.4573	11.6316	1000	1.7001	13.7131	7.1816	7.1631
1.425	17.4444	1500	1.7482	10.6395	5.9599	5.9460
1.4165	23.2573	2000	1.7743	10.9660	6.0619	6.0480
1.41	29.0702	2500	1.7913	11.3601	6.2961	6.2798
1.4195	34.8889	3000	1.8162	13.0714	8.0903	8.0741
1.4202	40.7018	3500	1.8379	14.6589	8.6073	8.5771
1.404	46.5146	4000	1.8379	12.0356	6.7829	6.7527
1.401	52.3275	4500	1.8736	13.8707	8.6397	8.6096
1.4006	58.1404	5000	1.8924	12.5760	7.2071	7.1723
1.4144	63.9591	5500	1.9047	12.4409	6.9614	6.9289
1.4142	69.7719	6000	1.9168	12.5760	7.0263	6.9915
1.4141	75.5848	6500	1.9243	12.6998	7.1167	7.0866
1.3999	81.3977	7000	1.9288	12.8912	7.2326	7.2025
1.3998	87.2105	7500	1.9306	12.9363	7.2419	7.2117
1.3999	93.0234	8000	1.9310	12.9363	7.2442	7.2141

Framework versions

• Transformers 4.48.3
• Pytorch 2.6.0+cu124
• Datasets 3.4.0
• Tokenizers 0.21.0