aliaamahgoub.github.io

Welcome

About Me

My name is Aliaa Mahgoub. I am a junior at Brooklyn Technical High School.

Projects

Sound Scene Classification

Coming soon! Please refer to my project Google Site for the time being.

Sound Event Localization and Detection

Experiment: Number of Mel bins for the Mel spectrogram

I experimented with the number of mel bins in the mel spectrogram to see how it affects the different evaluation metrics using both the microphone array (MIC) and first-order Ambisonics (FOA) formats. The original number of mel bins used in the model is 64.

Class-aware localization scores

nmel vs localization error nmel vs localization recall

Localization-aware detection scores

nmel vs error rate nmel vs f-score

Analysis

As expected, the FOA data resulted in better scores overall. However, the FOA and MIC formats had different optimal nmel numbers. The MIC data performed best overall with 256 Mel bins, while the FOA data performed best with 128 Mel bins.

Contact me

GitHub: AliaaMahgoub
Email: aliaamahgoub0@gmail.com
LinkedIn: Aliaa Mahgoub