herrschmidt0 / Audio-scene-classification

Thesis work. Research on audio scene classification and dimensionality reduction possibilities of time-frequency representations.

Clustering and Classification of Audio signals

Overview

Thesis work. Research on audio clustering and classification possibilities, focusing on the applicability of some dimensionality reduction methods.

Theory

The procedure consists of multiple steps:

• Feature extraction: conversion of samples to a time-frequency representation. Methods: STFT, log scaled STFT, mel scale, MFCC. Various frequency bin numbers are tested.
• Reshaping these spectrograms in a way that it is possible to perform dimensionality reduction on them.
• Applying dimensionality reduction. Project to 2D in case of clustering, and higher dimensions in case of classification. Methods: PCA, t-SNE, Isomap, SOM (Self Organizing Map).
• Classification or Clustering

Pipeline:

GUI

There's also a GUI implemented with PyQt5, that makes it possible to test and tune the methods described.

Papers

Some inspirational papers and articles are uploaded in the 'papers' folder.