mozilla-ai/whisper-large-v3-turbo-fi

Finetuned openai/whisper-large-v3-turbo on 2093 Finnish training audio samples from /home/kostis/cv-corpus-22.0-2025-06-20/fi.

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

Evaluation results on 1806 audio samples of Finnish:

Baseline model (before finetuning) on Finnish

• —Word Error Rate (Normalized): 16.576
• —Word Error Rate (Orthographic): 20.298
• —Character Error Rate (Normalized): 3.808
• —Character Error Rate (Orthographic): 4.498
• —Loss: 2.112

Finetuned model (after finetuning) on Finnish

• —Word Error Rate (Normalized): 11.972
• —Word Error Rate (Orthographic): 15.884
• —Character Error Rate (Normalized): 2.255
• —Character Error Rate (Orthographic): 3.119
• —Loss: 0.172