 |
This slide show the measured character and sentence accuracies.
The graphs of left-hand side and right-hand side are measured character and sentence accuracies respectively.
The conversion accuracy of task sets, report, numeral and proper noun are plotted for whisper models, base, small medium and large-v2.
As you can see, the character accuracy exceeded 90% for report and numeral tasks when the model is medium or larger one.
As for the sentence accuracy, the maximum conversion accuracy was around 80% at the maximum even when the model is medium or larger.
These results suggest we need a dedicated model for disaster response application.
|
IEEE/ICACT20250041 Slide.15
[Big Slide]