psst-model-4e-1s-hp600hz

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3383
• Wer: 0.1512
• Iu F1: 0.7329
• Iu Tp: 848
• Iu Fp: 468
• Iu Fn: 150

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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: 332
• training_steps: 4740

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Iu F1	Iu Tp	Iu Fp	Iu Fn
1.0822	0.4998	592	0.5258	0.2708	0.6667	398	220	178
0.9518	0.9996	1184	0.4797	0.2809	0.7484	461	195	115
0.5942	1.4989	1776	0.4785	0.2347	0.7359	386	87	190
0.5426	1.9987	2368	0.4593	0.2371	0.7721	476	181	100
0.2908	2.4981	2960	0.4965	0.2240	0.7621	458	168	118
0.2855	2.9979	3552	0.4893	0.2204	0.7719	467	167	109
0.0986	3.4973	4144	0.5502	0.2168	0.7612	435	132	141
0.0918	3.9970	4736	0.5491	0.2179	0.7733	452	141	124
0.0918	4.0	4740	0.5491	0.2182	0.7726	452	142	124

Framework versions

• Transformers 5.6.2
• Pytorch 2.6.0+cu124
• Datasets 2.21.0
• Tokenizers 0.22.2