Finetuned openai/whisper-tiny on 5 telugu training audio samples from example_data/custom.
This model was created from the Mozilla.ai Blueprint: speech-to-text-finetune.
Evaluation results on 5 audio samples of telugu:
Baseline model (before finetuning) on telugu
- Word Error Rate (Normalized): 207.368
- Word Error Rate (Orthographic): 219.149
- Character Error Rate (Normalized): 127.154
- Character Error Rate (Orthographic): 134.457
- Loss: 3.189
Finetuned model (after finetuning) on telugu
- Word Error Rate (Normalized): 15.789
- Word Error Rate (Orthographic): 100.0
- Character Error Rate (Normalized): 7.491
- Character Error Rate (Orthographic): 83.895
- Loss: 1.772