A new experimental technique for gathering similarity ratings for sounds

Abstract

Multidimensional scaling techniques (MDS) are a vibrant area of research with much development and advancement in last few decades. In this paper we focus on an interactive 2-dimensional interface for creating a similarity space for audio. Classical MDS techniques can place great demands upon participants overwhelming their sensory and cognitive abilities to make choices across such datasets using only pairwise comparison. We outline the results of an investigation to obtain multidimensional similarity ratings for a mixed sound collection containing both recorded sounds and synthesised sounds, and we describe briefly a computer application for collecting such data that reduces task demands by use of a 2-dimensional visual display with multistream audio for organising sound similarity and returns the results in the context of the whole stimuli set.