Finetuned openai/whisper-small on 450 Telugu training audio samples from /content/my_dataset.
This model was created from the Mozilla.ai Blueprint: speech-to-text-finetune.
Evaluation results on 51 audio samples of Telugu:
Baseline model (before finetuning) on Telugu
- Word Error Rate (Normalized): 183.319
- Word Error Rate (Orthographic): 188.0
- Character Error Rate (Normalized): 138.84
- Character Error Rate (Orthographic): 163.235
- Loss: 2.341
Finetuned model (after finetuning) on Telugu
- Word Error Rate (Normalized): 54.81
- Word Error Rate (Orthographic): 94.4
- Character Error Rate (Normalized): 30.103
- Character Error Rate (Orthographic): 38.904
- Loss: 0.693