ESSV Konferenz Elektronische Sprachsignalverarbeitung

Title: Phone mapping and prosodic transfer in speech synthesis of similar dialect pairs

Authors: Michael Pucher, Carina Lozo, Sylvia Moosmüller


In this paper we describe a phone mapping based method that can be used to synthesize a new dialect with an existing dialect model of a similar dialect. The method only uses transcriptions of original dialect data, which are then mapped onto the phones in the model. We use prosodic transfer of original duration and F0 to evaluate how the basic mapping model can be improved. We show that the prosodically enhanced models can outperform the basic model in a pairwise comparison task and can also achieve a slightly higher score on dialect authenticity. The goal of the proposed systems is to realize a dialect synthesis system with a small amount of symbolic training data that can come from transcribed dialect utterances or from the literature.

Year: 2017
In session: Sprachsynthese und regionale Varietäten
Pages: 180 to 185