Election Prediction with Automated Speech Emotion Recognition

Even though artificial intelligence has made meticulous outcomes, there is a gap and large thirst for speech emotion recognition in affective computing. Since machines don't understand human emotions, natural human-computer interaction is struggling a lot to strive and compete with today's market. The survey of past election results and prediction has become crucial and out-dated since the surveys could never clearly depicts the real mental state of an individual. Since the wealth and familiarity of the political people have a big title role in many surveys, it is not trustworthy. No surveys can understand the truth in the heart of the common people. So a novel approach is developed for day to day physical and psychological behavior of humans which is influenced by emotions. Other than behavioral expressions, emotion recognition from speech signals is a challenging modality. For the implementation, at least 100 audio signals of different areas, village, town, district, community, religion, and privileged people are considered and a nice questionnaire section (20 questions) is used to record the real emotion of the people. MFCC (mel-frequency cepstral coefficient) feature is extracted and classified with a decision tree, random forest, and CNN (convolutional neural network) to give the eight classes of emotions namely neutral, disgust, calm, angry, happy, sad, fearful and surprise. Any traditional algorithm can be compared to determine the performance of speech emotion recognition.