Back to list

Audiokodierung durch musikalische Fingerabdrücke

Authors Mur, T.
Year 2007
Thesis Type Diploma thesis
Topic Audio Signal Processing
Abstract This work aims to realize audio coding through musical fingerprints. Such fingerprints (thumbnails) can be representative pieces of a song alike chorus, verse, etc. or even smaller ones. If we find repeated parts in a song, these parts have to be saved just once and can be used manifold. In this way one gets a set of components and a construction plan for the song. First of all we have to divide the song in a meaningful way to calculate the similarity between each defined part. Therefore onset detection, i.e. Detection of the starting points for notes, or a beat tracking approach, i.e. Rhythm synchronisation, can be used to find start and end-points of each part. Within the next step, for each of these parts features will be extracted that submit a musically description. In this thesis spectral (MFCC's) and musical (Chroma; constant – Q) coefficients are investigated. Found these characteristics, repetitions are searched due to similarity – measurement (e.g. correlation). If highly correlated parts are found, a thumbnail and hence a basic component for the coding is located and stored. Furthermore the temporal locations of the detected similar parts within the song are stored in a look-up table, too. For each of these tasks several approaches are tested and compared.
Supervisors Sontacchi, A.