Comprehensive Approach to Multi Model Speech Emotion Recognition System

摘要

Human-Robot Interfaces have made it possible for robots to learn how to communicate with people and understand their feelings. Robotic dogs, for instance, should be able to comprehend not just verbal orders, but also other information, such as the emotional and physical state of its human commander, and adjust their behavior appropriately. The goal of this article is to learn how to recognize the feelings conveyed by a speaker's tone of speech. In today's world, detecting emotions has become an important task. When people speak in fearful, angry, or joyful emotions, their pitch range tends to be higher and broader, while their pitch range is lower when expressing these emotions. To learn both short-and long-term correlations in the log, a convolutional neural network (CNN) with four local feature-learning blocks and an LSTM layer can be utilised. This deep learning model is built using the Mel-spectrogram of the input audio recordings [1]. The audio characteristics MFCC, MEL, Chroma, and Tonnetz, as well as the Support Vector Machine and Multilayer Perception, were used in this study to identify different emotional states [4]. The audio files used by the algorithm are those found in the RAVDESS dataset, which is publicly available online. The system's effectiveness, which is the desired Outcome, will next be determined by analyzing the audio recordings' spectrograms in WAV format. [5].