Classify a persian speech dataset respect to their emotions using some classification models
The dataset I used is:
https://github.com/pariajm/sharif-emotional-speech-dataset
and you can download audio files and json file from this repo.
I used many diffrent classification models such as SVC, RandomForest, Knn, DecisionTree and MLP and best result I got was for MLP that it's accuracy was 74%.
These are visualisation of MLP results:
Of course it's not a very good accuracy because this dataset isn't really balanced which you can see that number of lables aren't equal and there are huge difrencess between them
v2- After I read some methodes about balancing data, I used one of this methodes to my data and I see a huge progress in accuracies.
here is number of voices respect to thier lables before balancing and after balancing:
Again I visualize MLP accuracy that is increased from 74 percent to 93 percent and confusion matrix below: