Cloning Dialects: Recreating and Localizing Dialectal Voices

Voice cloning technologies can replicate not only a speaker's voice but also potentially their regional dialect. This study evaluates the ability of the ElevenLabs model eleven multilingual v2 to preserve German dialects during voice cloning. Using dialectal recordings from nine regions, we generated cloned samples under varying synthesis conditions and assessed them through a perceptual localization task with human listeners. Results show that standard cloning tends to erase regional features, while style transfer and modified orthographic input partially improve dialect recognizability. The findings highlight the need for regionally informed models and further research into the acoustic and perceptual dimensions of dialect cloning.