An Automated System for the Assesment of Interview Performance through Audio & Emotion Cues

Automatic analysis and performance evaluation of interviewees is a largely unexplored and challenging problem. The proposed work provides a computational structure to enumerate the performance of the interviewee in the context of communication to give their performance feedback from the analysis of multimodal signals such as facial images and voice. A video captured during the interview is split into audio and visual frames. From the visual frames, the face is detected and their emotions are analyzed with the help of Histogram of Oriented Gradients. The facial expressions are classified by using Support Vector Machine. The classified facial expressions are happy, fear, sad, neutral, surprise, disgust and angry. From the audio cues,, the Mel Frequency Cepstral Coefficient features are extracted and this is categorized as fluent, non-fluent-pause and non-fluent-stammer. Both emotion and fluency of the candidate are fused to find performance score. This automated analysis provides the ratings for interview behavior such as poor, medium, high performance.