Exploring Emotional Signals: A Study on Audio-based Emotion Classification

Intersection of artificial intelligence and audio processing has witnessed a surge of interest in recent years, with various applications emerging, ranging from speech synthesis to music generation. This paper presents a comprehensive overview of various approaches to audio synthesis, utilising neural networks which classify the different emotions expressed in speech. It surveys a range of key areas such as speech classification, enhancement along with speaker dependent and speaker independent emotion analysis. By using an extensive dataset of emotional speech samples, neural network architecture effectively learns the sensitive nuances of emotional expression embedded within audio. The study delves into the utilization of various deep learning models such as long short-term memory (LSTM) networks, Auto-Encoders and convolutional neural networks (CNNs) which are capable of successfully capturing the nuanced emotional features within the input data, enabling the synthesis of emotionally enriched audio outputs. In this study, the model's capability to classify audio segments enriched with specific emotional content, such as happiness, sadness, anger, excitement etc is evidenced by the effectiveness of various approaches.