2D CNN Sound Classifier by Pew Pew Sounds

Oticon Audio Explorers Challenge 2023

The CNN architecture used in this implementation consists of two convolutional layers, each followed by a max-pooling layer and dropout for regularisation. After the convolutional layers, the output is flattened and passed through a fully connected layer with 64 units and a dropout layer. Finally, a softmax activation function is applied to the output layer, which consists of as many units as there are classes in the dataset.

Getting Started

Install python, we used version 3.10.11, and the packages listed below:

• numpy
• tensorflow
• scikit-learn

Installation

1. Clone the repo

   git clone https://github.com/andreaslborg/OticonChallenge2023

2. Install packages

   pip install numpy tensorflow scikit-learn

3. Run and train the model on all the data and predict on the test data

   python .\CNN2D_100.py

4. Run and train the model on 70% of the data and validate on 30%

   python .\CNN2D_70.py

Authors

• Alexander Løvig Borg
• Andreas Løvig Borg
• Anton Sig Egholm