whisper-audeering-classifier

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.9865
• Wer: 35.3575

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: 48
• eval_batch_size: 32
• 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: 18
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.07	0.1613	10	1.2246	37.8233
1.0062	0.3226	20	1.2108	37.7193
1.023	0.4839	30	1.1929	37.3268
1.0902	0.6452	40	1.1739	37.3872
1.0716	0.8065	50	1.1540	37.1255
0.9588	0.9677	60	1.1339	36.7732
1.0063	1.1290	70	1.1152	36.3673
0.9921	1.2903	80	1.0970	36.0553
0.9091	1.4516	90	1.0795	35.9211
0.9646	1.6129	100	1.0622	35.7601
0.8876	1.7742	110	1.0463	35.6594
0.8532	1.9355	120	1.0320	35.4447
0.8725	2.0968	130	1.0196	35.2870
0.8831	2.2581	140	1.0095	35.3675
0.8586	2.4194	150	1.0007	35.3776
0.8097	2.5806	160	0.9938	35.3004
0.8906	2.7419	170	0.9890	35.6561
0.8542	2.9032	180	0.9865	35.3575

Framework versions

• PEFT 0.16.0
• Transformers 4.57.3
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.22.2