git homework
using an old repo:
classifying audio based on the speaker's gender using Resnet50
By runnig PreProcess.m, some changes will be applied to the audio files (such as resizing) and finally they will be converted to Mel Spectograms and saved as images.
We have done the train procedure using MATLAB's Deep Network Designer APP's graphical interface. Due to our dataset's comparativly small size, we decided to use Transfer Learning: we loaded a pre-trained Resnet50 network and trained it again with our dataset. Finally we saved our trained model as trainedNetwork_1.mat file. Some data about the model and trainig process is provided in the trainInfoStruct_1.mat file.
You can simply load the trainedNetwork_1.mat file and test it. A sample script is provided in test.m.