Indoor environments are an essential part of the everyday context for everyone, influencing the health, well-being, and performance outcomes of occupants. Stuffy indoor air is a common problem caused by various factors, such as poor ventilation, occupant behaviour, and pollutants, which present challenges for automatic detection. Serving as an example study to understand the indoor context in an office environment, this study predicted perceived air stuffiness using multi-domain sensor data and self-reported labels from 23 knowledge workers, each for periods ranging from 3.5 to 7 months in natural settings. The study evaluated the applicability of three supervised machine learning methods: support vector, logistic regression, and random forest, in classifying knowledge workers' perceptions of air stuffiness at both personal and room levels. The results showed that machine learning models tailored to individuals and specific office rooms can predict “stuffy” and “not stuffy” perceptions, with random forest achieving the highest average accuracy of 69.6% for person-specific models and logistics regression achieving the highest average accuracy of 66.7%for room-specific models. The results suggest that a data-driven approach can help forecast perceived air stuffiness and potentially be used to optimise the indoor context to better support well-being and productivity in workplaces.

Emotions have a major role in our lives affecting on not only our experiences but also on our performance in different contexts. However, the physical environments often fail to read, interpret, and respond to the emotions of people present. The purpose of this study is to present a framework describing the needed technology modules and their relations to the surrounding environment in the context of adaptive spaces. The framework enables enhancement of physical spaces with emotionally aware, intelligent, and integrated services. For example, improving the experience of using a space or a service, detecting increased risk events, supporting vigilance, productivity, and well-being. The framework is built on studies focusing on two contexts with distinct characteristics: office and public transport vehicle. It aims to model the relationship and interaction between the human(s) and the environment facilitating the creation of adaptive services. The proposed framework consists of four main sections: measurement, modeling, actionable insight, and services. Finally, as the measurement of human behavior and emotional state often requires usage of sensitive personal data, issues related to the ethics of these solutions are discussed.

Cooking assistant is an application that needs to find a trade-off between providing efficient help to the users (e.g., reminding them to stir a meal if it is about to burn) and avoiding users' annoyance. This trade-off may vary in different contexts, such as cooking alone or in a group, cooking new or known recipe etc. The results of the user study, presented in this paper, show which features of a multimodal interface users perceive as socially acceptable or unacceptable in different situations, and how this perception depends on user's age.

In this paper, we describe an Augmented Reality Art installation, which reacts to user behaviour using Multimodal analysis of affective signals. The installation features a virtual tree, whose growth is influenced by the perceived emotional response from spectators. The system implements a 'magic mirror' paradigm (using a large-screen display or projection system) and is based on the ARToolkit with extended representations for scene graphs. The system relies on a PAD dimensional model of affect to support the fusion of different affective modalities, while also supporting the representation of affective responses that relate to aesthetic impressions. The influence of affective input on the visual component is achieved by mapping affective data to an L-System governing virtual tree behaviour. We have performed an early evaluation of the system, both from the technical perspective and in terms of user experience. Post-hoc questionnaires were generally consistent with data from multimodal affective processing, and users rated the overall experience as positive and enjoyable, regardless of how proactive they were in their interaction with the installation.

In this paper we present a novel light weight speaker clustering algorithm based on the Bayesian information criterion (BIC). Algorithm utilises BIC profiles, which were earlier used for false alarm compensation (FAC) in our speaker change detector (SCD). Proposed speaker segmentation followed by a light weight clustering is targeted to segment and label mobile device recordings directly in the device itself. Thus the main criterion in algorithm design was to maintain high detection accuracy while keeping computational costs in low level. Clustering algorithm gave F-score performance of 0.90 for speaker segmentation, which is 29% relative improvement compared to baseline results. Speaker segment labelling performance was 88%, when the number of speakers was undetermined. The experimental results indicate that our unsupervised speaker clustering algorithm is sufficiently effective and efficient for speaker segmentation applications in mobile devices.

In this paper we propose an unsupervised speaker change detection (SCD) system developed for mobile device applications. We use Bayesian information criterion (BIC) to find initial speaker changes, which are then verified or discarded in the second phase by utilizing modified BIC and silence detector information. Silence information usage after initial BIC in decision making is useful to separate real changes from noise peaks. Enhanced peak detector adjusts BIC penalty parameter automatically, which improve the robustness and feasibility. Improved BIC based false alarm compensation (FAC) merges effectively consecutive segments belonging to same speaker. Our experiments have shown the robustness of the algorithm and it produces very satisfactory results for difficult mobile phone recorded speech data.

Audio content analysis and automatic content management research field is facing new challenges from personal home video material created with mobile camera phones. We developed an audio content analysis and segmentation system that is robust to low sampling rate used in mobile phone camcorder tools and operates also well for AMR compressed data. Audio signal is segmented in five different classes, which can be used to classify videos for mobile video content management applications. The classification was done with Bayesian networks using topology of four-stage binary tree network that works in a hierarchical manner. Results were further smoothed within three second window using weighted sum of Bayesian networks class probabilities. The results show an average recognition accuracy of 88.6% for the high quality audio, 88.6% for the narrowband audio and accuracy of 82.9% for an AMR compressed audio. The effect of sampling rate to the recognition results is not significant and the effect of the AMR compression is also relatively low

Identifying users of portable devices from gait signals acquired with three-dimensional accelerometers was studied. Three approaches, correlation, frequency domain and data distribution statistics, were used. Test subjects (N=36) walked with fast, normal and slow walking speeds in enrolment and test sessions on separate days wearing the accelerometer device on their belt, at back. It was shown to be possible to identify users with this novel gait recognition method. Best equal error rate (EER=7%) was achieved with the signal correlation method, while the frequency domain method and two variations of the data distribution statistics method produced EER of 10%, 18% and 19%, respectively.