Researchers Launch Project to Generate Swiss Dialects Using AI

(CONNECT) According to a press release, the ZHAW School of Engineering and the Linguistics Center Zurich at UZH are investigating whether artificial intelligence (AI) can create voices that sound not just like a general region—such as Bern or Zurich—but like very specific towns. The focus is on dialect research, as CO-project leader Jan Milan Deriu from ZHAW explains: “Our vision is synthetic dialectology. We want to use knowledge from dialectology to improve language models—and, conversely, use synthetic data to answer questions in dialectology.”

The three-year project “Synthetic Dialectology of Swiss German” combines AI research and dialectology and has received funding from the Swiss National Science Foundation. Possible applications range from regionally tailored commercials and call centre voices to voice assistants and robotics. "We're already receiving inquiries from companies that want to use synthetic Swiss voices," said Deriu.

At the start of the project, the team is collecting voice recordings from three people in each of 85 towns in German-speaking Switzerland. Those in charge estimate that the process of recruiting participants and collecting data in the University of Zurich’s phonetics lab alone will take a year and a half.

At the same time, a proprietary voice-cloning model is being developed. It should not only reproduce well-known local dialects, but also be able to generate variants that do not appear directly in the data. As part of the evaluation, listening studies will examine how laypeople and experts perceive and geographically locate natural and artificial recordings. ce/mm