Articubench - An articulatory speech synthesis benchmark

Abstract:

The articubench benchmark is a collection of tasks, datasets and scores or metrics in the articulatory speech synthesis domain. The benchmark allows the comparison of different control models for the VocalTractLab speech synthesizer. Articubench is work in progress with the newest results and comparisons of models available at https://github.com/quantling/articubench. This publication states the motivation, goals and rationale behind articubench. Tasks, datasets and scores are presented.


Year: 2022
In session: Articulatory Synthesis
Pages: 43 to 50