Human Emotion Recognition System Using Deep Learning Approach

Satish Kumar .T, Mahesh .G, Amrutha.R, Arpitha.G, B. V.Divya, Bindu.B

Speech emotion recognition (SER) is gaining huge popularity in the recent days. SER not only covers the aspects of cultural background by utilizing its language for speech, but also by analyzing the emotions underlying it. This paper presents the available approaches of feature extraction, feature selection, classification and database. Various features namely,Mel Frequency Cepstral Co-efficients (MFCC), Linear-Prediction Cepstral Co-efficients (LPCC), Perceptual Linear Prediction (PLP) and Gammatone-Frequency Cepstral Co-efficients (GFCC) have been discussed. Also, the methods for classification such as Support Vector Machine (SVM), K-Nearest Neighborhood (K-NN), Principal Component Analysis (PCA) approach, Random forest etc., has been described.The databases like Berlin with 535 voices and RAVDESS with 2880 speech files and 2024 song files has been mentioned. In this paper, four emotions: Happy, Sad, Anger and Neutral are recognized using feed forward neural network. Multi- Layer Perceptron (MLP) and Long Short Term Memory (LSTM) are the methods utilized and the accuracies were found to be 58.53% and 72.68% respectively.

Volume 12 | 06-Special Issue

Pages: 907-915

DOI: 10.5373/JARDCS/V12SP6/SP20201109