Reusing Neural Speech Representations for Auditory Emotion Recognition

TL;DR

This paper explores reusing neural speech representations learned from large datasets to improve acoustic emotion recognition, achieving significant accuracy gains over traditional models.

Contribution

It introduces transfer learning architectures that leverage pre-trained neural speech representations for emotion recognition tasks.

Findings

~10% relative improvements in accuracy

Enhanced F1-score over baseline models

Demonstrates effectiveness of transfer learning in emotion recognition

Abstract

Acoustic emotion recognition aims to categorize the affective state of the speaker and is still a difficult task for machine learning models. The difficulties come from the scarcity of training data, general subjectivity in emotion perception resulting in low annotator agreement, and the uncertainty about which features are the most relevant and robust ones for classification. In this paper, we will tackle the latter problem. Inspired by the recent success of transfer learning methods we propose a set of architectures which utilize neural representations inferred by training on large speech databases for the acoustic emotion recognition task. Our experiments on the IEMOCAP dataset show ~10% relative improvements in the accuracy and F1-score over the baseline recurrent neural network which is trained end-to-end for emotion recognition.