What Is This Sound?
Project Brief
Wits (What Is This Sound?) is a Machine Learning project for Maynooth University CS401.
Introduction
The goal of this project is to use Neural Network to train a model for sound recognition in noisy envrironment (real-world environment). If the training is successful, we can use this model in a smartphone app for people with hearing loss to understand the sounds around them.
Methods
For this project, we will try to use Neural Network within TensorFlow to train a model.
Materials
This project will used Audio Dataset create by Google for the sound recognition in youtube video.
Expected results
A success for this project will be to recognize the differences between global category of sounds:
- noise
- voices
- animal
A more precise classification may be difficult to achieve.
Risk assessment
The main difficulty will be to transform the data for a suitable use in neural network. Two approachs will be try:
- Work with raw audio data
- Transform audio data in image data (in spectrogram for example)
We will try to compare the two methods.
Acknowledgments
This project is largely inspired by the work at Google Research (https://research.google.com/audioset/).
Statement
We will keep all the source code in git and commit and push often and arrange for you to have access to the repository.