Multidimensional Scaling as a Technique for Choosing Tone Signals

This paper reports a study that was conducted to select 3 or 4 tone signals from a set of 16 related signals with the criterion that the chosen signals sound as much alike as possible. The signals were required to be different for machine recognition purposes, but should sound alike to people to avoid possible confusion. The study employed a multidimensional scaling technique called INDSCAL to assess the similarity relationships among the 16 candidate signals. Twenty-two subjects provided similarity estimates on a 7-point scale for all possible pairs of signals. A 3-dimensional INDSCAL solution was found to describe the data adequately, and the required signals were chosen on the basis of their clustering in this space. The study demonstrates the use of multidimensional scaling to solve practical problems.