Korean articulatory speech synthesis using physical vocal tract model

Artificial vocal tract models provide the support of learning a second language and the therapy of speech disorders. Moreover, phonetic education and research can benefit from articulatory speech synthesis. Articulatory speech synthesis models are constructed by the source‐filter model of the human vocal tract. In this study, we generated a Korean articulatory speech synthesis model using Artisynth [Fels et al., ISSP, 419–426 (2006)], which is a 3‐D biomechanical open‐source simulation platform. As the origin of the Korean language, it has 10 basic vowel phonemes and 11 complicated vowels in which some vowels can be rounded and unrounded such as ∕eu∕, ∕yeo∕, ∕wae∕, etc. To synthesize these specific vowels, we created a new physical vocal tract model, which interconnects to form a complete integrated biomechanical system. The created model efficiently supports recording the Korean vowel sounds and linguistic analysis based on the linear prediction model. As a result, parameters of the glottis and controllable vocal tract filter are automatically evaluated. The acoustic quality of the synthesizer for Korean vowels is comparable with that of the existing commercial speech synthesis systems such as concatenation synthesizers [Donovan (1996)] and [Hamza (2000)].