Recognition of emotions present in human speech is a task made complicated due to incomplete knowledge of the relevant features, dependency on language, dialect and even individuals and the ambiguous meaning of term emotion itself. This work aims to study the relevant features present in human speech and infer its emotional category, making use of machine learning models. The prime objective of this work is to analyze the presented human speech and classify it into one of the seven emotional categories- happiness, sadness, anger, boredom, anxiety, disgust, and neutral. Proposed method extracts Mel Frequency Cepstral coefficients (MFCC), Chroma and Time Spectral features from Berlin EmoDB speech clips, and compares the results obtained from traditional classifiers to those with added nature inspired optimization technique. Egyptian Vulture Optimization Algorithm (EVOA), Grey Wolf Optimization (GWO) and Moth Flame Optimization (MFO) were applied over the mentioned feature set along with RF, KNN and SVM, and it is inferred that EVOA significantly improves the performance metrics of mentioned algorithms over EMO-DB when applied alongside. Using GWO with SVM produced the highest classification accuracy 90.67%.
Cite as: Sahu, S., Jain, A., Tiwari, R., Shukla, A. (2018) Application of Egyptian Vulture Optimization in Speech Emotion Recognition. Proc. 6th Workshop on Spoken Language Technologies for Under-Resourced Languages (SLTU 2018), 230-234, doi: 10.21437/SLTU.2018-48
@inproceedings{sahu18_sltu,
author={Shreya Sahu and Arpan Jain and Ritu Tiwari and Anupam Shukla},
title={{Application of Egyptian Vulture Optimization in Speech Emotion Recognition}},
year=2018,
booktitle={Proc. 6th Workshop on Spoken Language Technologies for Under-Resourced Languages (SLTU 2018)},
pages={230--234},
doi={10.21437/SLTU.2018-48}
}