Detecting protagonists and antagonists in the voice quality of American cartoon characters: a quantitative LTAS-based analysis

Abstract

Abstract The voices of heroes and villains in cartoons contribute to their uniqueness and helps shape how we perceive them. However, not much research has looked at the acoustic properties of character voices and the possible contributions these have to cartoon character archetypes. We present a quantitative examination of how voice quality distinguishes between characters based on their alignment as either protagonists or antagonists, performing Principal Component Analysis (PCA) on the Long-term Average Spectra (LTAS) of concatenated passages of the speech of various characters obtained from four different animated cartoons. We then assessed if the categories of “protagonists” and “antagonists” (determined via an a priori classification) could be distinguished using a classification algorithm, and if so, what acoustic characteristics could help distinguish the two categories. The overall results support the idea that protagonists and antagonists can be distinguished by their voice qualities. Support Vector Machine (SVM) analysis yielded an average classification accuracy of 96% across the cartoons. Visualisation of the spectral traits constituting the difference did not yield consistent results but reveals a low-versus-high frequency energy dominance pattern segregating antagonists and protagonists. Future studies can look into how other variables might be confounded with voice quality in distinguishing between these categories.