yezarniko/pharmacy-whisper-en

<!-- 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. -->

Pharmacy ASR Whisper Base Model

This model is a fine-tuned version of openai/whisper-base.en on the Medicines ASR Dataset dataset yezarniko/medicines which contains real-world drug names and pharmacy-friendly pronunciations. It achieves the following results on the evaluation set:

• —Loss: 0.0008
• —Wer: 7.7151

Model description

The goal of this model is to provide fast and accurate transcription of medicinal terms, improving speech recognition for healthcare and pharmacy applications.

Intended uses & limitations

• —Medicine search systems
• —Prescription reading
• —Voice-based pharmacy automation
• —Speech-driven drug inventory/apps

The base model is Whisper-base.en, meaning it only supports English. It cannot recognize medicine names spoken in other languages. The dataset may not represent all drug names available in real pharmacies.

Training and evaluation data

yezarniko/medicines

• —Total Dataset: 14669
• —Total Duration (second)s: 15853.571
• —No. of Agents: 10
• —No. of VoiceLines: 1466

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 1e-05
• —trainbatchsize: 16
• —evalbatchsize: 8
• —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: 500
• —training_steps: 5000
• —mixedprecisiontraining: Native AMP

Training results

Framework versions

• —Transformers 4.50.3
• —Pytorch 2.6.0+cu124
• —Datasets 2.14.4
• —Tokenizers 0.21.2