Rahul3215/whisper-small-telugu

Finetuned openai/whisper-small on 812 Telugu training audio samples from /content/dataset.

This model was created from the Mozilla.ai Blueprint: speech-to-text-finetune.

Evaluation results on 203 audio samples of Telugu:

Baseline model (before finetuning) on Telugu

• —Word Error Rate (Normalized): 173.255
• —Word Error Rate (Orthographic): 139.928
• —Character Error Rate (Normalized): 142.764
• —Character Error Rate (Orthographic): 161.132
• —Loss: 2.305

Finetuned model (after finetuning) on Telugu

• —Word Error Rate (Normalized): 122.977
• —Word Error Rate (Orthographic): 134.42
• —Character Error Rate (Normalized): 83.015
• —Character Error Rate (Orthographic): 104.111
• —Loss: 1.345