Evaluation of Synthesised Sound Effects

Abstract

The current fi eld of sound synthesis research presents a range of methods and approaches for synthesising a given sound. Sounds are synthesised to facilitate interaction or control of a sound, to enable sound searching through parametric control of a sound or to allow for the creation of an arti ficial nonexistent sound. In all of these cases, the ability of a synthesis technique to reproduce a desired sound is integral. This thesis uses an audio feature representation of audio to produce a sonically inspired taxonomy, based entirely on the sonic content of sound, which enables a user to search through a large set of sounds without the need for understanding of context. This provides an approach for using audio features to compare similarity between different audio effect samples in a sound effects library. This thesis then develops approaches for evaluation of synthesised sound effects. A large scale methodic subjective evaluation of synthesised sound effects is performed, evaluating a range of different synthesis methods in a range of different sound classes or sonic contexts. It is then identi fied that there are cases where synthesised sound effects can be considered as realistic as a recorded sample. An objective evaluation approach is then presented. Audio feature vectors are used to measure the relative objective similarities between two samples, and this is correlated with a perceptual evaluation of sound similarity. These objective measures are then compared based on the perceptual evaluations. Both evaluation approaches are then demonstrated in a case study of aeroacoustic sound effects, where these subjective and objective evaluation techniques are demonstrated for a speci fic case. There is no single best approach to synthesising sound effects. More consistent and rigorous evaluation methodologies will lead to a better understanding as to the advantages and disadvantages of each method. The outcome of this research suggests that further consistent perceptual and objective evaluation within the sound effect synthesis community will lead to a better understanding as to the successes and failings of existing work and thus facilitate an enhancement of current sound synthesis technologies.