Document Type

Theses, Ph.D

Rights

This item is available under a Creative Commons License for non-commercial use only

Disciplines

Electrical and electronic engineering

Publication Details

Sucessfully submitted for the award of Doctor of Philosopy (Ph.D) to the Dublin Institute of Technology, 2010.

Abstract

For centuries, luthiers have committed to working towards better understanding and improving the sound characteristics and playability of violins. With advances in technology and signal processing, studies attempting to define a violin’s sound quality via physical characteristics and resonance patterns have ensued. Existing work has primarily focused on physical aspects reflecting and instrument’s sound quality. In the music information retrieval domain, advances have been made in areas such as instrument identification tasks. Although much research has been completed on finding suitable features from which musical instruments can be represented, little work has focused on the violin’s complete timbre space and the effect a player has on the sound produced. This thesis specifically focuses on representing violin timbre such that a computer can detect the sound associated with a beginner from that of a professional standard player and detect typical beginner playing faults based on analysis of waveform signal only. Work has been limited to nine playing faults considered by professional musicians to be typical of beginner violinists. In order to achieve these goals, it was necessary to create a suitable dataset consisting of an equal number of beginner and professional standard legato note samples. Feature extraction was then carried out by taking features from the time, spectral and cepstral domains. Selected features were then used to represent the samples in a classifier based on their efficacy at reflecting change within the violin’s timbre space. The dataset underwent the scrutiny of professional standard stringed instrument players via listening tests from which the target audience’s perception was captured. This information was verified and normalised before use as a priori labels in the classifier. Based on different feature representations, classification of violin notes reflecting perceived sound quality is presented in this thesis. The results show that it is possible to get a computer to determine between beginner and professional standard player legato notes and to detect playing faults. This thesis involves a thorough understanding of violin playing, its perception, suitable analysis methods, feature extraction, representation and classification.