The traditional face-to-face interview is the most common recruitment interviews. However,Besides being time-consuming,it can be difficult for some interviewers to read candidates because there is no way to know what their voice sounds like under stress or how to properly grade an applicant based on how they responded.In contrast with Virtual interviews that don't take place in person.The interviewer within short period of time judges the entire personality of the candidate only basis of the answers given by the interviewee.Our framework takes audio as an input and predicts the interview score followed by feedback about candidates personal and emotional traits based on prosodic features and lexical model. By using prosodic features our framework considers emotions like anger, happiness, anxiety and sorrow.While our lexical model works on the major personality characteristics.There will be a marked psychological traits such neuroticism and extraversion. RAVDESS Emotional Speech and Song data set is used for audio analysis.To train our lexical model we used a Stream-of-consciousness data set. Therefore, in our project we use time distributed CNN along with LSTM for audio analysis and NLP with RNN is used to build lexical model.For final score prediction we choose Decision Tree Regression Model.Finally, our framework is less susceptible to human biases and thus improving the interviewers ability to make a hiring decision

.