Feasibility study on synthesis of English vowels with a vocal tract mapping interface

Abstract

This study describes the synthesis of English vowels with a vocal tract mapping interface developed in our laboratory. The interface used five vocal tract shapes located at the apexes of a pentagonal chart in consideration of Japanese vowels. It allowed users to configure a vocal tract shape corresponding to the location of a mouse pointer based on interpolation. Furthermore, it performed an inverse estimation of a vocal tract from formant frequencies. Although the interface was initially developed for Japanese vowels, whether English vowels can also be synthesized with this interface is an interesting topic for confirmation of its versatility. Therefore, we investigated the feasibility of producing English vowels ([α], [æ], [ɔ], [ε], [ɝ], [i], [I], [u], [℧], [∧]) with the mapping interface. First, we investigated whether the mapping interface could estimate vocal tract shapes for English vowels synthesized with the VowelEditor of Praat software as test sounds. Furthermore, listening tests were conducted to examine whether the re-synthesized vowel sounds with the mapping interface matched the original vowel sounds of Praat. The experimental results showed that the mapping interface has potential for producing English vowels.