Experiments with one-class classifier as a predictor of spectral discontinuities in unit concatenation

Abstract

We present a sequence of experiments with one-class classification, aimed at examining the ability of such a classifier to detect spectral smoothness of units, as an alternative to heuristics-based measures used within unit selection speech synthesizers. A set of spectral feature distances was computed between neighbouring frames in natural speech recordings, i.e. those representing natural joins, from which the per-vowel classifier was trained. In total, three types of classifiers were examined for distances computed from several different signal parametrizations. For the evaluation, the trained classifiers were tested against smooth or discontinuous joins as they were perceived by human listeners in the ad-hoc listening test designed for this purpose.