Emotion Recognition in Human-Computer Interaction
How can we make human – computer interactions more effective and easier?
Final-year BTech students of Department of Electronics and Communication Engineering, Abhijith N. Kashyap, Siddhartha Kumar and Visarad S. are attempting to do just that through their project Speaker Emotion Recognition for Human-Computer Interaction.
Funded by the Karnataka State Council for Science and Technology (KSCST), the project is guided by Dr. Shikha Tripathi, Chairperson, Departments of Electronics and Communication Engineering and Electrical and Electronics Engineering.
Automatic emotion recognition in speech, during the past decade, has shifted from being one of tangential concern, to a major topic in the area of human-computer interaction and speech processing.
“Our aim is to enable a very natural interaction with the computer, through speech instead of use of traditional input devices such as keyboard, mouse or even a touch-screen. We want to have the machine understand not only the verbal commands, but also subtle emotional cues that might accompany what is said,” explained the students.
The team first developed an algorithm to classify speech content into various categories differentiated by emotions of anger, boredom, disgust, fear, happiness or sadness. There is even a ‘neutral’ category.
“The speech emotion recognition system we designed, consisted of three principal parts for signal processing, feature calculation and classification. It was after an intensive literature survey on various existing emotion recognition algorithms that we proposed our algorithm that should enable better feature selection and recognition rate,” they added.
“Signal processing involves digitalization and potentially acoustic pre-processing like filtering, as well as segmenting the input signal into meaningful units. Feature calculation is concerned with identifying relevant
features of the acoustic signal with respect to emotions such as Pitch, Formant Frequency, Energy and Mel Frequency Cepstral Coefficients (MFCC), etc. Classification, lastly, maps feature vectors onto emotion classes through learning by examples, using Fuzzy ART Map Neural Network (FAMNN) classifier,” the students further explained.
The results obtained after the initial tasks were completed were satisfactory and the team now plans to implement the algorithm on a Digital Signal Processor (DSP) for real-time testing. The algorithm will also be incorporated in a service robot, in the future, to have it interact with emotion in its verbal commands.
The technology, if further developed, can also potentially find application in diverse areas such as, emotionally aware in-car systems, tutoring systems that adapt pedagogical strategies to a student’s emotional state, etc.
The team was happy on receiving funding for their project and developing the algorithm successfully. Now, as they plan to implement, we congratulate them and wish them all the best.
August 15, 2013
School of Engineering, Bengaluru