Read it to me: An emotionally aware Speech Narration Application

TL;DR

This paper explores emotional style transfer in speech audio using MelGAN-VC and evaluates the emotional accuracy with an LSTM classifier, highlighting better results for sadness due to its consistent expression.

Contribution

It introduces a method for emotional style transfer in speech audio using MelGAN-VC and assesses emotional transfer quality with an LSTM classifier.

Findings

Sad emotion transfer is more successful than happy or anger.

Generated sad audio closely matches human expressions of sadness.

The approach demonstrates potential for emotionally aware speech narration.

Abstract

In this work we try to perform emotional style transfer on audios. In particular, MelGAN-VC architecture is explored for various emotion-pair transfers. The generated audio is then classified using an LSTM-based emotion classifier for audio. We find that "sad" audio is generated well as compared to "happy" or "anger" as people have similar expressions of sadness.