whisper-accent-small.en-accent-head-only

This model is a fine-tuned version of openai/whisper-small.en on the westbrook/English_Accent_DataSet dataset. It achieves the following results on the evaluation set:

• Loss: 0.4207
• Accent Accuracy: 0.8668

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• 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: linear
• lr_scheduler_warmup_steps: 0.05
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accent Accuracy
No log	0	0	3.1399	0.0106
2.3869	0.1271	200	2.1690	0.2847
2.3640	0.2541	400	2.1340	0.1863
2.2865	0.3812	600	1.8785	0.2201
1.9448	0.5082	800	1.5942	0.3446
1.4611	0.6353	1000	1.2579	0.5222
1.2292	0.7623	1200	1.0262	0.6351
1.2505	0.8894	1400	0.8094	0.7239
0.8404	1.0159	1600	0.6767	0.7606
0.8986	1.1429	1800	0.6125	0.7963
0.8090	1.2700	2000	0.5430	0.8195
0.7181	1.3970	2200	0.5028	0.8417
0.5592	1.5241	2400	0.4625	0.8398
0.5376	1.6512	2600	0.4207	0.8668
0.6433	1.7782	2800	0.3927	0.8658
0.4718	1.9053	3000	0.3781	0.8668

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2