Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

Curriculum learning for speech emotion recognition from crowdsourced labels

Abstract: 

This study introduces a method to design a curriculum for machine-learning to maximize the efficiency during the training process of deep neural networks (DNNs) for speech emotion recognition. Previous studies in other machine-learning problems have shown the benefits of training a classifier following a curriculum where samples are gradually presented in increasing level of difficulty. For speech emotion recognition, the challenge is to establish a natural order of difficulty in the training set to create the curriculum. We address this problem by assuming that ambiguous samples for humans are also ambiguous for computers. Speech samples are often annotated by multiple evaluators to account for differences in emotion perception across individuals. While some sentences with clear emotional content are consistently annotated, sentences with more ambiguous emotional content present important disagreement between individual evaluations. We propose to use the disagree- ment between evaluators as a measure of difficulty for the classification task. We propose metrics that quantify the inter- evaluation agreement to define the curriculum for regression problems and binary and multi-class classification problems. The experimental results consistently show that relying on a curriculum based on agreement between human judgments leads to statistically significant improvements over baselines trained without a curriculum.

up
0 users have voted:

Paper Details

Authors:
Reza Lotfian, Carlos Busso
Submitted On:
20 May 2020 - 9:43am
Short Link:
Type:
Presentation Slides
Event:
Presenter's Name:
Reza Lotfian
Document Year:
2020
Cite

Document Files

Slides of paper

(34)

Subscribe

[1] Reza Lotfian, Carlos Busso, "Curriculum learning for speech emotion recognition from crowdsourced labels", IEEE SigPort, 2020. [Online]. Available: http://sigport.org/5408. Accessed: Aug. 10, 2020.
@article{5408-20,
url = {http://sigport.org/5408},
author = {Reza Lotfian; Carlos Busso },
publisher = {IEEE SigPort},
title = {Curriculum learning for speech emotion recognition from crowdsourced labels},
year = {2020} }
TY - EJOUR
T1 - Curriculum learning for speech emotion recognition from crowdsourced labels
AU - Reza Lotfian; Carlos Busso
PY - 2020
PB - IEEE SigPort
UR - http://sigport.org/5408
ER -
Reza Lotfian, Carlos Busso. (2020). Curriculum learning for speech emotion recognition from crowdsourced labels. IEEE SigPort. http://sigport.org/5408
Reza Lotfian, Carlos Busso, 2020. Curriculum learning for speech emotion recognition from crowdsourced labels. Available at: http://sigport.org/5408.
Reza Lotfian, Carlos Busso. (2020). "Curriculum learning for speech emotion recognition from crowdsourced labels." Web.
1. Reza Lotfian, Carlos Busso. Curriculum learning for speech emotion recognition from crowdsourced labels [Internet]. IEEE SigPort; 2020. Available from : http://sigport.org/5408