Basil Preisig

Supervisor(s): Basil Preisig

How are artificially generated "deepfake" voices detected?

Description: AI-powered voice cloning now makes it possible to produce realistic copies of human voices from just a few seconds of audio. Although such technologies - often referred to as "deepfakes" - are increasingly present in everyday life, a large part of the population is barely aware of their existence.

The main goal of this master's thesis is to investigate whether people can distinguish AI-generated voices from natural ones, and which acoustic features — such as prosody or pitch — are particularly relevant in doing so. The findings are intended to inform the development of materials and tools that help people better protect themselves in everyday life from deceptive content such as fake phone calls.

Number of theses available: 1

Timeline: By arrangement

Contact: basil.preisig@uzh.c

Brief description

Prerequisite scientific, technical knowledge, programming skills...