Speech and Emotion Recognizer Using Deep Learning and Librosa

Speech and Emotion Recognition (SER) means to identify the underlying emotion by extracting the features of a human from his/her voice. Emotion recognition is a part of speech recognition. SER has many applications in today’s world. Its applications can be found in healthcare, education, security, telecommunication industries, Human Computer Interaction (HCI), etc. Through this analysis it can be found whether a customer is unhappy or satisfied. SER is an open research area and there are many algorithms available which classify these voice signals and detect the emotions in it. Also what features influence the detection or change in a voice are yet uncertain. Here in this paper, techniques like Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and a feature named Mel-frequency Cepstrum Coefficient (MFCC) are used. Various python libraries to extract the emotion out of the speech signal provided are utilized. Various datasets are used namely Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) and Surrey Audio-Visual Expressed Emotion (SAVEE).