GoranS/whisper-large-v3-turbo-hr-parla

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

whisper-large-v3-turbo-hr-parla

This model is a fine-tuned version of openai/whisper-large-v3 on the classla/ParlaSpeech-HR dataset and additional 400h private dataset, both with augmented additions.

It achieves the following results on the evaluation set:

• —Loss: 0.0816
• —Wer: 3.52%

WER comparsion

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 6.25e-06
• —trainbatchsize: 64
• —evalbatchsize: 32
• —seed: 42
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —lrschedulerwarmup_steps: 800
• —num_epochs: 2
• —mixedprecisiontraining: Native AMP

Training results

Framework versions

• —Transformers 4.46.3
• —Pytorch 2.5.0+cu121
• —Datasets 3.1.0
• —Tokenizers 0.20.3