On the Use of Vocal-Tract Approximations for Instrumental Quality Assessment

The nature of linear prediction becomesmost vivid through the relation between predictor and reflection coefficients in the context of the source-filter model of speech production. In particular, an approximation of the vocal-tract shape can be recursively calculated using the reflection coefficients only. To date, it is not fully clear to what extent such an approximation can be exploited for a perception-related quality analysis of connected speech signals. In this paper, we will review known approaches from literature and elaborate on a novel extension towards instrumental quality assessment of synthesized speech signals.