Publications
Novel Speech Features of Disturbances
in Laryngeal Muscle Coordination
Branimir Dropuljić, Davor Petrinović, Ćosić, Krešimir
Original scientific paper published in Proceedings of 7th IEEE International Conference on Cognitive Infocommunications (CogInfoCom 2016), Wroclaw, Poland, 16-18.10.2016.
The present paper proposes speech features derived from the fundamental frequency (F0) contour decomposition. The decomposition method is designed in order to differentiate, as much as possible, simultaneous neurobiological effects on vocal fold vibration. The focus of this paper is placed on involuntary disturbances of such vibrations, which are analyzed in the context of emotional stress. The proposed features are compared with conventional perturbation measures, i.e. jitter and shimmer, using two datasets: Synthetic perturbations and SUSAS (Speech Under Simulated and Actual Stress) subset – Roller- coaster. Features are additionally analyzed in the context of elimination potential of voluntary effects like F0 contour changes during natural pronunciation. Results of the initial synthetic perturbation analysis indicate that the proposed features could be less affected by the voluntary control and, on the other hand, more related to disturbances in laryngeal muscle coordination. The proposed features generally outperform conventional perturbation features in speech under stress analysis.
Learn more
Analyzing Affective States
using Acoustic and Linguistic Features
Branimir Dropuljić, Sandro Skansi, Robert Kopal
Original scientific paper published in Proceedings of Central European Conference on Information and Intelligent Systems (CECIIS), Varaždin, Hrvatska, 21-23.09.2016.
This paper explores the hypothesis that sentiment in text is closely related to emotions in speech in terms of features needed for successful detection. We use a Croatian emotional speech corpus (CrES) and a Croatian social network textual sentiment corpus SentHR. We first perform emotional state estimation based on acoustic speech features using support vector machines in the first case and random forest in second. Accuracy between 60% and 70% was achieved for five discrete emotion classification task. Subsequently, we trained a positive naive Bayes classifier for textual sentiment, reporting an accuracy of around 70% (with a pronounced bias towards the complement). Finally, we used the trained sentiment classifier for two classification experiments on the transcripts of the CrES dataset for classifying anger and sadness. Across several iterations, the results showed that accuracy on the transcripts was around 50% for both sadness and anger, reporting a slightly higher (albeit consistently higher) accuracy on emotional state "anger".
Learn more
Why social physics is
The Next Big Thing?
Robert Kopal
Professional paper published in Mreža in August/September 2016.
How to take advantage of social learning to find the best ideas? How to increase the call center efficiency by 23%, while at the same time reducing the stress for 19%? How to reduce power consumption by 17%? How to detect depression and other psychological or mental disorders of people based on their behavioral characteristics on social networks and the Internet? How to automatically recognize emotional states of users during the dialogue with the operator in a call center? Or how to find out in real time the users’ engagement for the service that the operator offers? How to recommend music and movies to users based on a mood they cause? The answer to those questions and many others provide social physics.
Learn more
Reciprocal Payers Identification
in Banking Logs using SAT Solvers
Sandro Skansi, Branimir Dropuljić
Original scientific paper published in IEEE Mipro 2016 Proceedings, Opatija, Croatia, 30.5.-3.6.2016.
In this paper we presented solvers for satisfiability testing (SAT) as a novel approach to finding reciprocal payers in banking logs. A term "reciprocal payers" is usually treated as general fraud by using standard techniques such as expert systems, machine learning and in recent times social network analysis. SAT as a technique for data analysis was abandoned due to the unfeasibility of SAT solvers. SAT solvers, however continued to develop in the hardware and software verification communities. We presented a proof-of- concept solution for identification of reciprocal payers (formally called a clique), which is a group of bank clients that issue payments to each other (each member to each member). We do not use real data due to client confidentiality, but the reader can see the principle. In the basic approach it is assumed that each client has only one account, and in the extended, second approach, it was allowed that a client can have more than one account.
Learn more
Social network analysis
Practical approach
Robert Kopal, Darija Korkut, Saša Krnjašić.
Algebra & IN2data, 2016.
Networks are all around us. By applying structured analytical techniques, we could say that there is still enormous network potential that is underused. The main goal of this book is to describe how social network analysis can be used for discovering non-trivial knowledge and structures of groups and teams in various fields of industry and society in general.
Learn more