Speech Emotion Recognition Using Deep Learning

Speech Emotion Recognition (SER) involves identifying and analyzing emotions conveyed through speech signals. It utilizes techniques from machine learning and signal processing to extract features from audio data, enabling the classification of emotions such as happiness, sadness, anger, fear, disgust, and neutrality. SER has applications in human-computer interaction, virtual assistants, mental health monitoring, and sentiment analysis. The paper employs two distinct models for speech emotion recognition. The first model utilizes Long Short-Term Memory (LSTM) networks, starting with data preprocessing via Librosa, including resampling and feature extraction such as MFCCs. Following dataset splitting, the LSTM architecture, optimized for sequential data, incorporates input layers, multiple LSTM layers, and dense layers with suitable activation functions. Model training involves monitoring progress using validation set metrics. The second model explores real-time emotion recognition using live audio input, employing Convolutional Neural Networks (CNNs) and LSTM Networks. Data preprocessing, feature extraction, and selection are executed before model training. The findings contribute to advancing speech-emotion recognition systems, with potential applications in human-computer interaction, virtual assistants, and mental health monitoring. The study identifies and categorizes seven primary emotions within speech signals: Anger, Disgust, Neutral, Fear, Happy, Sad, and Surprise, providing a comprehensive framework for understanding and interpreting emotional cues in human speech.