Thesis

Back to list

Sound Event Detection for Smart Cars

Authors Linke, J.
Year 2018
Thesis Type Master's thesis
Topic Audio Signal Processing
Keywords Music Information Retrieval, Neuronal Networks
Abstract This thesis designs classification models from the area of artificial intelligence to distinguish between urban noise and siren sounds. Audio recordings from the urban environment and audio recordings capturing specific sound events are collected in a basic data set. With audio signal processing suitable audio features for four classification approaches are extracted and selected from the obtained data environment. Different machine learning classification algorithms are discussed for two classification tasks: Two approaches for a binary classification task and one approach for a classification task with three classes are presented. The last approach compares the best binary classification solution with a deep learning classifier in the sense of transfer learning. All classification models are tested with a self-recorded validation set including car microphone recordings from the urban environment and the so-called Martinshorn.
Supervisors Sontacchi, A.