chengyili2005/whisper-medium-DINA

<!-- 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 Medium — English / Spanish / Miami Bangor

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 24.0 (English) — gender-balanced subset, the Common Voice 24.0 (Spanish) — gender-balanced subset and the Bangor Miami Corpus — code-switching Spanish/English interviews, segmented at utterance level from CHAT transcripts datasets. It achieves the following results on the evaluation set:

• —Loss: 0.2052
• —Wer: 6.8973
• —Cer: 2.5627

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: 5e-05
• —trainbatchsize: 4
• —evalbatchsize: 4
• —seed: 42
• —gradientaccumulationsteps: 4
• —totaltrainbatch_size: 16
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —lrschedulerwarmup_steps: 500
• —training_steps: 122820

Training results

Framework versions

• —PEFT 0.18.1
• —Transformers 4.52.0
• —Pytorch 2.9.1+cu128
• —Datasets 4.5.0
• —Tokenizers 0.21.4