Katrin Schoenenberg

Same but different? — Using speech signal features for comparing conversational VoIP quality studies

In this paper we demonstrate how speech signal features can be used to detect and explain differences in human to human conversation tests. To this end, we compare the results of two conversational VoIP quality experiments designed to quantify the impact of network delay on perceived speech quality. Both studies followed the same procedures and used the same scenarios, but were conducted in two different labs. Our comparison shows that the two studies, despite having been executed correctly using the same test design, still can produce surprisingly different results regarding the users quality perception on a MOS scale. In this respect, speech signal features extracted from conversation recordings help identifying divergent participant behavior as plausible cause for such differences. Our in-depth analysis reveals how novel parameters developed by us like Intended and Unintended Interruption Rate (IIR, UIR) and the corrected Speaker Alternation Rate SARcorr can be used to successfully determine the extent to which the results of different conversational speech quality studies are directly comparable and thus eligible for pooling, or not.

On interaction behaviour in telephone conversations under transmission delay

Abstract This work analyses the interaction behaviour of two interlocutors communicating over telephone connections affected by echo-free delay, for conversation tasks yielding different speed and structure. Based on a series of conversation tests, it is shown that transmission delay in a telephone circuit does not only result in a longer time until information is exchanged between the interlocutors, but also alters various characteristics of the conversational course. It was observed that with increasing transmission delay, the realities perceived by the interlocutors increasingly diverge. As a measure of utterance pace, a new conversation surface structure metric, the so-called utterance rhythm (URY), is introduced. Using surface-structure analysis of conversations from different conversation tests, it is shown that peoples’ utterance rhythm stays rather constant in close-to-natural conversations, but is considerably affected for scenarios requiring fast interaction and a clear answering structure. At the same time, the quality of the connection is perceived less critically in close-to-natural than in tasks requiring fast interaction, that is, interactive tasks leading to a delay-dependant utterance rhythm. Hence, the conclusion can be drawn that the degree of necessary adaption of the utterance rhythm to a certain delay condition co-determines the extent to which transmission delay impacts the perceived integral quality of a call.

Why are you so slow? – Misattribution of transmission delay to attributes of the conversation partner at the far-end

Abstract This study addresses the question of how transmission delay affects user perception during speech communication over telephone systems. It aims to show that the occurrence of pure delay should not be neglected when planning a telephone or conferencing system even if no impact on the perceived quality of the call can be found. It is, for instance, known that, the communication surface structure changes dramatically when transmission delay is inserted by the communication system. Furthermore, studies suggest a change in the perception of the interlocutor at the far-end. This paper describes two experiments that assess the misattribution of the technical impairment delay to personality and behavior-related attributes of the conversation partners. The first experiment shows that interlocutors are perceived as being less attentive when conversing in a three-party setting with symmetrical and asymmetrical delay conditions. In the second experiment, the misattribution is considered in more detail looking at ascribed personality attributes in two-party interaction under transmission delay. For both experiments, comparing the conversation surface structure of delayed to non-delayed calls helped to understand the found outcomes.

Multimedia Conferencing and Telemeetings

With today’s technical possibilities, in particular, packet-based data transmission and high processing power, telephone and videoconferencing systems celebrate increasing interest. However, the success of such systems is essentially determined by the quality provided and experienced when using them. This is why a high need of appropriate assessment methods can currently be observed. Given the broad range of possible solutions, assessing QoE of so-called telemeetings becomes very difficult and brings along the need for a high degree of variability regarding assessment methods. Since multiple participants usually communicate via such systems, it is required to also investigate aspects of the interaction process and their influence on QoE. Furthermore, the multiparty situation enables users to directly perceive asymmetries in the equipment and in qualities provided from different sites, which affects the perceptual situation as well. This chapter is intended to explain the described challenges in detail and to give first insights into how they might be handled.

Quality of Experience and Interactivity

This chapter discusses the relation between interactivity and QoE. In this context, a definition of interactivity comprising human-to-human interaction as well as human-to-machine interaction is presented, and a description of a possible instrumentation is given. In terms of quality formation, a mediation layer between quality influence factors and perceived quality features is introduced that allows the inclusion of interactivity-related perception in the quality formation process. A discussion of commonalities and differences between interaction with a system and interaction with one or several other persons via a system identifies the open challenges for reliable and successful measurement of interactivity related aspects and the identification of relationships between these interaction measures and QoE.

Methods for Human-Centered Evaluation of MediaSync in Real-Time Communication

In an ideal world people interacting using real-time multimedia links experience perfectly synchronized media, and there is no latency of transmission: the interlocutors would hear and see each other with no delay. Methods to achieve the former are discussed in other chapters in this book, but for a variety of practical and physical reasons, delay-free communication will never be possible. In some cases, the delay will be very obvious since it will be possible to observe the reaction time of the listeners modified by the delay, or there may be some acoustic echo from the listeners’ audio equipment. However, in the absence of echo, the users themselves do not always explicitly notice the presence of delay, even for quite large values. Typically, they notice something is wrong (for example “we kept interrupting each other!”), but are unable to define what it is. Some useful insights into the impact of delay on a conversation can be obtained from the linguistic discipline of Conversation Analysis, and especially the analysis of “turn-taking” in a conversation. This chapter gives an overview of the challenges in evaluating media synchronicity in real-time communications, outlining appropriate tasks and methods for subjective testing and how in-depth analysis of such tests can be performed to gain a deep understanding of the effects of delay. The insights are based on recent studies of audio and audiovisual communication, but also show examples from other media synchronization applications like networked music interaction.

A conversation analytic approach to the prediction of leadership in two to six-party audio conferences

When evaluating the Quality of Experience (QoE) of audio conferences the participants talking behaviour is critical for their quality judgement. A dominant person, for instance, who talks and interacts a lot is likely to be more disturbed by conversation-related impairments than a less active person. Social interaction analysis provides reliable nonverbal audio features to predict dominance in face-to-face meetings. However, little is known about pure audio conferences. This is why this study investigates common dominance features for audio-only conferences. In the first part, we look at different group sizes and whether prediction models exist that predict the leader for two-to six-party audio-only conversations. In a second step, we apply the best model to an unknown audio conferencing dataset. Finally, we tested if dominance features are associated to the assigned caller or a particular person in each group and how the callers talking behaviour is related to perceived quality.