stdbug/whisper-tiny-ba

Finetuned openai/whisper-tiny on 133675 Bashkir training audio samples from mozilla-foundation/commonvoice17_0.

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

Evaluation results on 14513 audio samples of Bashkir:

Baseline model (before finetuning) on Bashkir

• —Word Error Rate (Normalized): 150.765
• —Word Error Rate (Orthographic): 127.801
• —Character Error Rate (Normalized): 116.224
• —Character Error Rate (Orthographic): 115.431
• —Loss: 5.831

Finetuned model (after finetuning) on Bashkir

• —Word Error Rate (Normalized): 102.544
• —Word Error Rate (Orthographic): 103.049
• —Character Error Rate (Normalized): 89.277
• —Character Error Rate (Orthographic): 89.293
• —Loss: 1.441