The model does not generalize well
#1
by
Daijek
- opened
I built a model for this dataset also but I did it from scratch in mine with mfcc features. The issue that I have with mine which I can also see is evident in yours is that when ever the model hears an accent that is not from the dataset, it predicts welsh, would you have any idea why this happens or maybe how to fix it
Actually no. I do not have any idea and I experienced this myself as well.
Have you created a completely different model, yet it predicts welsh?
I built several models, actually. I built a CNN, an LSTM and I even explore some transfer learning with VGGish and Yamnetand yet the issue still persists