How Intense Are Your Words? Understanding Emotion Intensity from Speech

Himadri Mukherjee, Hanan Salam, Alice Othmani, K. C. Santosh

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Speech-based emotion recognition has emerged as an important research area in the field of affective computing. Despite the multifarious developments in emotion recognition, the analysis of the intensity of the expressed emotion has mostly been unexplored. As compared to distinguishing emotions, understanding their intensities is a herculean task. The spectral characteristics of a speech signal portray important information which is necessary to distinguish emotional intensity. Line Spectral Frequency (LSF) features offer a spectral representation of the speech signal in addition to modeling the formant structure. Such features have been explored for modeling emotions from speech. However, they have not been explored for emotion intensity detection. In this paper, we explore LSF features for emotion intensity characterization. In order to make the proposed approach appropriate for low computational resources settings, we present a low dimensional version of LSF: the Band Dominance and Dynamics LSF (BDD-LSF). The proposed BDD-LSF is based on a frequency bands dominance and dynamics analysis technique of the LSF features. Such feature is capable of handling intra-class variation between different intensities involving multifarious emotional states. Using the publicly available RAVDESS dataset, we achieved the highest accuracy of 75.75% for distinguishing emotional intensities. Our system also outperforms reported works which use deep learning-based techniques.

Original languageEnglish (US)
Title of host publication2021 IEEE 21st International Conference on Communication Technology, ICCT 2021
PublisherInstitute of Electrical and Electronics Engineers Inc.
Number of pages7
ISBN (Electronic)9781665432061
StatePublished - 2021
Event21st IEEE International Conference on Communication Technology, ICCT 2021 - Tianjin, China
Duration: Oct 13 2021Oct 16 2021

Publication series

NameInternational Conference on Communication Technology Proceedings, ICCT


Conference21st IEEE International Conference on Communication Technology, ICCT 2021


  • Affective Computing
  • Line Spectral Frequency
  • Random Forest
  • SER
  • Speech Signal Processing
  • Speech-based Emotional Intensity Recognition

ASJC Scopus subject areas

  • Electrical and Electronic Engineering
  • Computer Networks and Communications


Dive into the research topics of 'How Intense Are Your Words? Understanding Emotion Intensity from Speech'. Together they form a unique fingerprint.

Cite this