MAGEFACE: Performative Conversion of Facial Characteristics into Speech Synthesis Parameters

Abstract

In this paper, we illustrate the use of the MAGE performative speech synthesizer through its application to the conversion of realtime-measured facial features with FaceOSC into speech synthesis features such as vocal tract shape or intonation. MAGE is a new software library for using HMM-based speech synthesis in reactive programming environments. MAGE uses a rewritten version of the HTS engine enabling the computation of speech audio samples on a two-label window instead of the whole sentence. Only this feature enables the realtime mapping of facial attributes to synthesis parameters.