Emotion Detection from Audio: A Deep Learning Approach
Keywords:
Convolutional Neural Network (CNN), Deep Neural Network (DNN), Human-computer interaction, Long Short Term Memory (LSTM), Speech Emotion Recognition (SER), Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS)Abstract
Human expressions and emotions play a vital role in Emotional Recognition. Universally, emotions can be often categorized and grouped into mere types like anger, happiness, sadness, fear, surprise, disgust and even neutral emotional state. Speech Emotion Recognition (SER) has been gaining increasing popularity in the growing field of Human–Computer Interactions (HCIs). The Speech Emotion Recognition (SER) system is used to extract and predict the emotional tone of a speaker through audio signals. Emotion detection is one of several things that can be inferred from an audio signal, including age, gender, person, and action. A difficult issue in audio signal processing is emotion recognition from speech or audio. Human speech is recognised for carrying linguistic information in addition to the identity and feelings of the speaker. Emotional recognition from speech or audio is a challenging problem in audio signal processing. This proposed method deals with the Emotion detection from Audio Speech using CNN and LSTM methods. This method of emotion recognition is based on the spectrograms and mel-spectrograms that are returned from short voice recordings. An adequate model was required to learn and analyse sequential data, such as audio files. In this proposed method, a long short-term memory Recurrent Neural Network (RNN) model is utilised. The primary goals are to create a system capable of identifying speech-embedded emotions and to make accurate predictions.
