A Study on Voice Actor Recommendation for Game Characters Based on Acoustic Feature Estimation and Document Co-occurrence
Erika Sakai, Takayuki Itoh, Akinori Ito · 2017
Animation and game are very popular in Japan. Many works in animation and game have been released, and many characters have been produced. It is an important factor to select appropriate voice actors so that we can appropriately impress the characters. Based on this idea, we are developing a prototype of voice actor recommendation tool. We recorded voice performance from portable games and calculated vocal acoustic features. Then, we conducted a questionnaire regarding impressions of the recorded voice performance, and applied SOM (Self-Organizing Map) to solve the relationship between acoustic features and impression values. We can estimate appropriate acoustic features of new characters when we specify their impression values, and consequently generate a list of appropriate voice actors for these new characters based on their acoustic features. At the same time, we collected documents describing these characters on the Web, and calculated distances among them by a natural language processing tool. We can discuss how to assign voice actors to the game characters by observing the list of voice actors and distance-based character map. Since our development is still an on-going work, this paper presents visualization results to demonstrate the appropriateness of the technical components in the presented tool.