Evaluation methods for dialect speech synthesis of similar dialect pairs
* Presenting author
Dialect synthesis is a challenging area of research and contrasts the synthesis of standard varieties not only as to the non standard nature of dialects but also in collecting proper corpus data. Previously we evaluated a method for synthesizing new dialects with existing dialect models of a similar dialect by using a simple phone mapping. Then we used a small amount of training data to transfer the original duration and fundamental frequency (F0) of a speaker in order to evaluate how the basic mapping model can be improved.In this contribution we focus on the evaluation methods of synthesized dialects. To improve dialect synthesis we should not only adapt the existing acoustic models but also the evaluation methods. It is expected that the presentation of synthesised dialect to the listener is crucial to the rating of these systems. Due to the versatile connotations of dialects we assume that a sterile evaluation setting seems inappropriate to the listener and needs to meet the situative demands. In the evaluation process we ask the listener to envision scenarios where they e.g. need to make a telephone call to a local district information line or order a taxi.