Anticipatory coarticulation in predictive articulatory speech modeling

Abstract:

The aim of the present study is to test whether effects of anticipatory coarticulation emerge from a segment based and a recurrent gradient based planning speech resynthesis in the articulatory speech synthesizer VocalTractLab. For this, natural articulations of /baba/, /babi/ and /babu/ are recorded using ultrasound. While anticipatory coarticulation is observable in the articulatory movements and the phonetic signal of the human recording, these patterns are not observable in the segment based resynthesis, and partially observed in the recurrent gradient based planning resynthesis. The recurrent gradient based planning shows anticipatory coarticulation in formant shifts, but not in tongue raising.


Year: 2021
In session: Postersession 2
Pages: 208 to 215