SynthDiaSG

Can computer voices sound like “proper” Swiss German, and can we calibrate them in a way that they represent the actual speech of a specific town or village (instead of just a generic form of Swiss German with a few regional traits)? The SynthDiaSG project develops speech synthesis for local Swiss dialects. At the same time, the project uses artificial voices as a new tool for dialect research.

Modern speech synthesis can imitate voices with remarkable realism, but so far it treats Swiss dialects only as broad regional varieties such as “Bernese German” or “Zurich German.” In reality, however, dialects often differ from one place to another. The project has two main aims. First, we develop speech synthesis models that can reproduce dialects at a local level and smoothly blend between neighboring localities. Second, through listening experiments, we investigate whether non-experts and specialists can recognize and locate natural and synthetic dialect recordings and whether they accept them as valid representations of the corresponding dialects. This gives rise to “Synthetic Dialectology”: a new approach that makes artificial voices usable as a controllable instrument for dialect research. By the end of the project, we can give informed answers to questions such as which linguistic features are crucial for the perception of dialects and whether synthetic voices can replace real recordings in research studies.

Dialects are a central part of the identity of German-speaking Switzerland. Language technology that reflects local variation can enable more natural voice assistants and accessible applications, and the methods can be transferred to other languages with regional diversity. A collaboration between ZHAW and the University of Zurich, the project brings together speech technology and dialectology. Data and results will be made publicly available; on dialekterkennung.ch, the public will be able to explore dialect diversity for themselves.

SynthDiaSG (2026–2029) is an interdisciplinary project hosted by the Centre for Artificial Intelligence, School of Engineering ZHAW and the Zurich Center for Linguistics, University of Zurich and funded by the Swiss National Science Foundation (grant number 10007532).

Co-PIs

Prof. Dr. Mark Cieliebak (Computer Science)
Prof. Dr. Anja Hasse (Linguistics)

Postdocs

Dr. Jan Milan Deriu (Computer Science)
Dr. Patrick Mächler (Linguistics)

PhD Student

N.A.

Student Assistants

BA Corinne Utzinger
BA Julia Weller