whisper-tiny-en-minds14

This model is a fine-tuned version of openai/whisper-tiny on the PolyAI/minds14 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5909
• Wer Ortho: 0.3288
• Wer: 0.3241

Model description

openai/whisper-tiny

Intended uses & limitations

automatic-speech-recognition

Training and evaluation data

PolyAI/minds14

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
5.6090	0.8929	25	1.9933	0.4775	0.3867
1.0683	1.7857	50	0.5234	0.3812	0.3701
0.6299	2.6786	75	0.4771	0.3553	0.3424
0.4853	3.5714	100	0.4679	0.3405	0.3335
0.3242	4.4643	125	0.4824	0.3584	0.3489
0.1566	5.3571	150	0.4964	0.3325	0.3235
0.1353	6.25	175	0.5071	0.3374	0.3264
0.0663	7.1429	200	0.5246	0.3436	0.3335
0.0493	8.0357	225	0.5306	0.3387	0.3300
0.0299	8.9286	250	0.5481	0.3362	0.3306
0.0163	9.8214	275	0.5651	0.3294	0.3253
0.0076	10.7143	300	0.5702	0.3307	0.3259
0.0093	11.6071	325	0.5824	0.3307	0.3253
0.0058	12.5	350	0.5861	0.3263	0.3217
0.0054	13.3929	375	0.5874	0.3307	0.3253
0.0045	14.2857	400	0.5902	0.3294	0.3247
0.0048	15.0	420	0.5909	0.3288	0.3241

Framework versions

• Transformers 5.12.1
• Pytorch 2.11.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2