| Abstrakt: |
Human emotion recognition from audio has potential applications such as healthcare, feedback assessment, gaming and advertisement to mention few. Human emotion detection often helps in assessing the feelings of person automatically. It could lead to making well informed decisions. Advancements in machine learning (ML) has paved way for unprecedented possibilities in emotion recognition from audio automatically. The methods identified in existing literature exhibit limitations, notably the absence of feature engineering to enhance predictive performance. A framework is proposed based on ML for automatic recognition of human emotions from a given voice content. The framework is called the Human Emotion Recognition Framework (HERF), designed to receive audio datasets as input and employs supervised learning for the automated identification of human emotions based on audio signals. We proposed two algorithms for realizing the framework. A Hybrid Feature Selection (HFS) algorithm is introduced to enhance the efficiency of identifying features that could have discriminative power. Additionally, the Neural Network-based Automatic Emotion Recognition (NN-AER) algorithm, utilizing Multilayer Perceptron (MLP) and HFS, is proposed for automatic emotion recognition. The feature selection provided by the HFS algorithm improves the training quality of NN-AER. RAVDESS is dataset used for empirical study. This dataset supports emotions such as neutral, happy, sad, disgust, angry, fearful and surprised. We designed a web application used to recognise emotion for given audio sample based on saved MLP model. Results of our study revealed that NN-AER outperforms many states of the art methods. [ABSTRACT FROM AUTHOR] |
