Vocal entrainment is a social adaptation mechanism in human interaction, knowledge of which can offer useful insights to an individual’s cognitive-behavioral characteristics. We propose a context-aware approach for measuring vocal entrainment in dyadic conversations. We use conformers (a combination of convolutional network and transformer) for capturing both short-term and long-term conversational context to model entrainment patterns in interactions across different domains. Specifically we use cross-subject attention layers to learn intra- as well as interpersonal signals from dyadic conversations. We first validate the proposed method based on classification experiments to distinguish between real (consistent) and fake (inconsistent/shuffled) conversations. Experimental results on interactions involving individuals with Autism Spectrum Disorder (ASD) also show evidence of a statistically-significant association between the introduced entrainment measure and clinical scores relevant to symptoms, including across gender and age groups.

Diagnostic procedures for ASD (autism spectrum disorder) involve semi-naturalistic interactions between the child and a clinician. Computational methods to analyze these sessions require an end-to-end speech and language processing pipeline that goes from raw audio to clinically-meaningful behavioral features. An important component of this pipeline is the ability to automatically detect who is speaking when i.e., perform child-adult speaker classification. This binary classification task is often confounded due to variability associated with the participants' speech and background conditions. Further, scarcity of training data often restricts direct application of conventional deep learning methods. In this work, we address two major sources of variability-age of the child and data source collection location-using domain adversarial learning which does not require labeled target domain data. We use two methods, generative adversarial training with inverted label loss and gradient reversal layer to learn speaker embeddings invariant to the above sources of variability, and analyze different conditions under which the proposed techniques improve over conventional learning methods. Using a large corpus of ADOS-2 (autism diagnostic observation schedule, 2nd edition) sessions, we demonstrate up to 13.45% and 6.44% relative improvements over conventional learning methods.

This paper presents a novel fuzzy based approach to gesture driven human robot interaction. Now a day, gestures are considered to be the most effective communicative medium for remotely controlling a robot. In this work, the gestures are employed to instruct a Khepera II robot to move from a specific starting position to a goal position following a specific path. The main highlight is the determination of exact degree of rotation with proper application of acceleration and brake in order to reach the specified goal position without hitting the obstacles. A Takagi-Sugeno-Kang based fuzzy model with two antecedents (type-2 fuzzy sets) and one consequent (crisp value) has been employed to determine the angle of rotation. The performance of the proposed framework has been tested in terms of a number of parameters like accuracy, precision, error rate etc. And in each case, the formulated strategy has proved its worth.

Currently, gesture recognition from continuous video sequences is one of the most exciting research areas. This paper proposes a novel HMM-based gesture recognition scheme that can be implemented for developing an improved HCI system capable of providing enhanced performance. This framework explores the high potential of Microsoft's Kinect sensor in gesture recognition by utilizing it in the data acquisition phase. The primary novelty of the work lies in the choice of an active difference signature-based feature descriptor that contains time-warped information in a single sequence over the classically used geometric features. The discussed framework has been tested for 12 distinct gestures embodied by 60 different subjects and it is important to note that for all the gestures the proposed scheme has attained a fairly high recognition rate of nearly 90% which proves the worth of the present work in real time applications. Further, to check the efficacy of the newly formulated framework the performance of the same has been validated against the existing standard technologies.

This paper proposes a novel evolutionary approach to the optimal selection of electrodes as well as relevant EEG features for effective classification of cognitive tasks. The problem has been formulated in the framework of a single objective optimization problem with an aim to simultaneously satisfy three criteria. The first criterion deals with maximization of the correlation between the features of EEG sources before and after the selection of the optimal electrodes. The second criterion is concerned with minimization of the mutual information between the features of the selected EEG electrodes. The last criterion aims at maximization of the ratio of the difference between the selected features of the EEG sources between and within any two cognitive tasks. A self-adaptive variant of firefly algorithm is proposed to solve the above optimization problem by proficiently balancing the trade-off between the computational accuracy and the run-time complexity. Experiments undertaken over wide variety of cognitive tasks reveal that the proposed algorithm outperforms the other standard algorithms (applied to the same problem) in terms of accuracy and computational overhead.

With the growing interest in the domain of human computer interaction (HCI) these days, budding research professionals are coming up with novel ideas of developing more versatile and flexible modes of communication between a man and a machine. Using the attributes of internet, the scientists have been able to create a web based social platform for learning any desired art by the subject himself/herself, and this particular procedure is termed as electronic learning or e-learning. In this paper, we propose a novel application of gesture dependent e-learning of dance. This e-learning procedure may provide help to many dance enthusiasts who cannot learn the art because of the scarcity of resources despite having great zeal. The paper mainly deals with recognition of different dance gestures of a trained user such that after detecting the discrepancies between the gestures shown and actually performed by a novice; the user can rectify his faults. The elementary knowledge of geometry has been employed to introduce the concept of planes in the feature extraction stage. Actually, five planes have been constructed to signify major body parts while keeping the synchronous parts in one unit. Then four distances and four angular features have been obtained to provide entire positional information of the different body joints. Finally, using a probabilistic neural network the dance gestures have been classified after training the said network with sufficient amount of data recorded from numerous subjects to maintain generality.