whisper-SER-base-v7(skip_special_tokens=True during and lr = 1e-05 steps = 12k ,warmup = 500)

This model is a fine-tuned version of openai/whisper-base on the Whisper_Compatible_SER_benchmark + enhanced_facebook_voxpopulik_16k_Whisper_Compatible dataset. It achieves the following results on the evaluation set:

• Loss: 0.0978
• Wer: 56.9573

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 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: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3141	0.5510	1000	0.3218	42.8881
0.1626	1.1019	2000	0.2021	58.5652
0.1553	1.6529	3000	0.1462	87.1676
0.1091	2.2039	4000	0.1199	63.8528
0.1069	2.7548	5000	0.1027	63.3271
0.042	3.3058	6000	0.0958	66.8831
0.0434	3.8567	7000	0.0935	77.2418
0.0254	4.4077	8000	0.0926	64.4712
0.0265	4.9587	9000	0.0939	59.9876
0.0136	5.5096	10000	0.0955	58.2870
0.009	6.0606	11000	0.0985	62.9561
0.0067	6.6116	12000	0.0978	56.9573

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.2
• Tokenizers 0.21.0