Abteilung Digitale Phonetik

Our research interests span various areas, including speech, natural language processing, machine learning, and human-machine interaction. These can be divided into the following four categories: Perception, Interaction, Learning, and Reasoning.

Among multiple senses (e.g., hearing, vision, and touch) in Perception, our research focuses on the sub-area of processing and extracting information from speech. We aim to improve the robustness and fairness of speech processing systems, such as speech recognition and speech emotion recognition, in real-world conditions, including spontaneous conversations, noisy environments, and multilingual settings (e.g., Code-switching).

Interaction research focuses on methods to teach systems which strategy to apply for intelligently reacting to user inputs (from both lexical inputs and social signals), which linguistic realization should be used (even across languages), and how to utter it (e.g., via controllable speech synthesis) to increase human acceptance. One of the main objectives of this area is to make systems adaptive in supporting users to efficiently fulfill specific goals, while also being friendly and likable by using the right words and tones.

Learning research focuses on methods to equip systems with the ability to learn quickly (e.g., using meta-learningtechniques) and to learn continually. Thus, it enables faster fine-tuning that is closer to human learning than the traditional supervised learning framework, and continually improves in the case of data distribution shifts (e.g., in fake audio detection).

Reasoning research investigates methods that give systems the capability to provide trustworthy outputs (i.e., ideally no hallucinations, bias-free) and ideally an explanation along with their decisions. We focus on identifying supporting facts rather than logical reasoning. Furthermore, we are also interested in methods that allow systems to estimate and communicate reliable uncertainty values for their decisions.

Our research group adheres to the ethical guidelines provided by various research organizations, including ISCA, ACL, and IEEE, particularly in the context of artificial intelligence (AI) systems.