manohar2405/whisper-small-telugu-manohar

Finetuned openai/whisper-small on 99 telugu training audio samples from /content/drive/MyDrive/1781705178557-cv-corpus-26.0-2026-06-12-te/cv-corpus-26.0-2026-06-12/te.

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

Evaluation results on 93 audio samples of telugu:

Baseline model (before finetuning) on telugu

• —Word Error Rate (Normalized): 211.709
• —Word Error Rate (Orthographic): 197.708
• —Character Error Rate (Normalized): 174.02
• —Character Error Rate (Orthographic): 229.888
• —Loss: 2.799

Finetuned model (after finetuning) on telugu

• —Word Error Rate (Normalized): 73.466
• —Word Error Rate (Orthographic): 102.292
• —Character Error Rate (Normalized): 44.25
• —Character Error Rate (Orthographic): 61.264
• —Loss: 1.08