Citation

Haron, Anis and Wong, Chee Onn and Hew, Soon Hin (2022) Deriving Quantitative Aural Aesthetics: Towards a Quantifiable Measure of Timbre. In: Creative Multimedia Postgraduate Research Colloquium, 14 July 2022, Multimedia University, Malaysia.

Text Proceedings of the FCM PGRC 2022.pdf - Published Version Restricted to Repository staff only Download (942kB)

Abstract

The aim of this thesis is to formulate and propose an approach for a quantifiable measure of timbre. A quantitative measure for timbral description introduces granularity compared to the common practice of using semantic descriptors. Results from our research finds potential in applications such as parameter auto-recommendation systems and as a basis for ranking mechanisms. We asked what are the most commonly used semantic descriptors for description of tone colour, what are the key features required for a formulation of a quantifiable measure of timbre and what is the viability of our proposed approach. We do so with a descriptive research design approach paired with quantitative observation methods and numerical analysis. Additionally, a normative survey was conducted and results are compared to test the viability of our proposed approach. We categorised our research into three distinct stages. These stages are identification, classification and viability test. The identification stage answers the following question, what is the best fit semantic descriptor for this sample? We approached this stage using spectral energy audio features and hidden Markov models. While the classification stage answers the following question, based on the identified best fit semantic descriptor, what is the intensity of the description for this sample? This stage was approached using spectral centroid, spectral complexity and dissonance audio features. A normalised weighted sum score were calculated from the feature extraction to arrive at a final value that represents a quantitative measure for the timbre in question. Finally, the viability test stage answers the following question, are the results obtained from identification and classification stages correct? A survey was conducted where participants were asked to identify and classify pairs and groups of randomly selected audio samples. Responses that form the majority in the conducted survey were considered as normal. Survey responses were then compared against results obtained from our proposed identification and classification approach. We found that our proposed approach for a quantifiable measure of timbre yields relative measure values. Additionally, for certain identified descriptors, larger measure values denote higher intensity. While for other identified descriptors, smaller measure values denote higher intensity. Comparison of our proposed identification and classification approach against the results of the conducted normative survey are highly in line, suggesting that our proposed approach is highly viable.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Timbre, Audio Descriptors, Quantitative Measure of Timbre
Subjects:	N Fine Arts > NX Arts in general
Divisions:	Faculty of Creative Multimedia (FCM)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	24 Nov 2022 03:56
Last Modified:	24 Nov 2022 03:57
URII:	http://shdl.mmu.edu.my/id/eprint/10768

Downloads

Edit (login required)