Skip to content

MJSahebnasi/IE_hw1

Repository files navigation

IE_hw1

git homework

using an old repo:

Deep-Gender-Detector-for-Audio

classifying audio based on the speaker's gender using Resnet50

Preprocess

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.

Train

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.

train stats

Test

You can simply load the trainedNetwork_1.mat file and test it. A sample script is provided in test.m.



More details are provided in this document. (written in Persian)

About

git homework

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages