Presenter

Mrs Selma Yilmazyildiz - ETRO, Vrije Universiteit Brussel [Email]

Abstract

Recent developments in robotics, artificial intelligence, and machine learning are further accelerating the introduction of robots in our daily lives and the physical environment around us. In order for humans and robots to co-habit in a common space, robots must behave and operate in ways that are similar to or acceptable by humans. In essence, they also need to be social, as we are.
The design, development and study of these social robotic agents and their interactions with humans form up the young but growing field of social Human-Robot Interaction (sHRI). The social robotic agents are equipped with social cues ranging from the use of bodily and facial gestures, natural language and eye gaze to more unique and robot specific methods such as expression through colors, synthetic sounds and vocalizations
This thesis introduces the umbrella concept of Semantic-Free Utterances (SFU) and brings together multiple sets of studies in social HRI that have never been analyzed jointly before. SFUs are composed of vocalizations and sounds without semantic content or language dependence that may still facilitate rich communication and expression during sHRI. Currently they are most commonly utilized in animation movies (e.g., WALL-E), cartoons (e.g., “Teletubbies,”), and computer games (e.g., The Sims) and hold significant potential for applications in sHRI.
In this thesis, a Semantic-Free Affective Speech (SFAS) Framework, which allows robots to express and communicate through vocalizations of meaningless strings of speech sounds (also referred to as affective gibberish speech), has been developed. This framework provides a complete set of tools that can be used as a vocal communication medium for an agent and allows to study diverse aspects of affective human-robot interaction.
As a component of this SFAS framework, a semantic destruction technique that allows a given intelligent text in a certain language to turn into semantic-free gibberish text that is still natural sounding has been developed. Using the methods and techniques outlined, an emotional gibberish speech database (EMOGIB) has been built and made available to the HRI community for further research.
SFAS framework was further enhanced with two modification techniques that are instrumental to utilize the framework across various scenarios in social HRI. One of them is the voice modification capability which provides the alignment of the voice characteristics of the gibberish speech voice with the robot morphology. The second modification, a concatenative synthesis approach which is referred to as segment swapping, decreases the cost of implementation of the framework in HRI studies which will hopefully lead to wider and faster adoption of the framework by the HRI community.
Piloting the implementation of the outlined Semantic-Free Affective Speech (SFAS) Framework, sets of experiments that assess the effectiveness of using SFAS across various aspects of affective human-robot interaction were performed. The results of these experiments have shown the expansive applicability of the proposed framework in social HRI, while outlining certain improvement areas in various components used in the pilot implementations.

Short CV

Master in Applied Computer Science, VUB, Belgium, 2006