Emotion Recognition from Mizo Speech: A Signal Processing Approach

As humans, the most natural way for expressing feelings and communicating is verbal speech. Communication with machines is done either by traditional input methods i.e., keyboard, joystick or with the help of speech recognition technology. The speech recognition mostly works on an average input of user but it lags the instant response, based on the mood of the user and the emotional feature of communication. Emotion recognition is one of the most popular and expanding research domain in recent years. The world is moving towards the functioning of machines with less manual input. Human-computer interaction can be improved by emotions recognition, reducing the need for human intervention. The paper presents an emotion recognition system developed by utilizing the different machine learning techniques that helps to process and classify speech signals to detect emotions. The input to the system are speech audios that were recorded in the Mizo language, the native language of Mizoram, India. To develop the emotion recognition system, seven basic emotions i.e., Anger, Disgust, Fear, Happiness, Neutral, Sadness and Surprise, were analyzed from emotional speech signals. Further, to extract the features from audio signals, signal processing methods have been used. In this work twelve features were extracted, those are Autocorrelation, Chroma, Fundamental frequency, Zero-crossing rate, Short-time energy, RMS, Periodogram, LPCC, MFCC etc. The extracted features were used to train four classifiers namely Decision tree, KNN, SVM and Ensemble. At last, a model was prepared with the combination of classifiers that recognizes the emotion of the signal. The emotion recognition system recognizes single emotions as well as mixed types of emotions. The insights gained may be helpful in a wide range of applications.