arielcerdap/whisper-medium-fluencybank

Whisper fine-tuned on FluencyBank — openai/whisper-medium

This model is a fine-tuned version of openai/whisper-medium on the FluencyBank Timestamped dataset. It achieves the following results on the evaluation set:

• Loss: 1.8983
• Wer: 15.9086
• Cer: 10.9154

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: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 2500
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.4549	11.6279	250	1.7186	11.7776	6.6503
1.4261	23.2558	500	1.7611	10.8548	6.2588
1.4204	34.8837	750	1.8104	10.7888	6.2679
1.4216	46.5116	1000	1.7901	10.9207	6.4819
1.4179	58.1395	1250	1.8390	10.9426	6.4637
1.4168	69.7674	1500	1.8682	15.7328	10.7515
1.4164	81.3953	1750	1.8841	15.9086	10.8517
1.4161	93.0233	2000	1.8941	15.8207	10.8790
1.416	104.6512	2250	1.8984	15.9525	10.9882
1.416	116.2791	2500	1.8983	15.9086	10.9154

Framework versions

• Transformers 4.45.2
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.20.3