Lightweight Deep Learning Framework for Speech Emotion Recognition

Speech Emotion Recognition (SER) system, which analyzes human utterances to determine a speaker’s emotion, has growing impact on how people and machines interact. Recent growth in human-computer interaction computational intelligence drawn the attention of many researchers Artificial Intelligence (AI) deep learning because its wider applicability several fields, including computer vision, natural language processing, affective computing, among others. Deep models do not need any form manually created features they can automatically extract prospective from input data. models, however, call for lot resources, high processing power, hyper-parameter tuning, making them unsuitable lightweight devices. In this study, we focused developing an efficient model speech emotion recognition with optimized parameters without compromising performance. Our proposed integrates Random Forest Multi-layer Perceptron(MLP) classifiers into VGGNet framework recognition. The was evaluated against other based methods (InceptionV3, ResNet, MobileNetV2, DenseNet) it yielded low complexity optimum experiment carried out three datasets TESS, EMODB, RAVDESS, Mel Frequency Cepstral Coefficient(MFCC) were extracted 6-8 variants emotions namely, Sad, Angry, Happy, Surprise, Neutral, Disgust, Fear, Calm. demonstrated performance 100%, 96%, 86.25% accuracy RAVDESS respectively. This revealed that achieved higher compared recent state-of-the-art found literature.