Emotional End-to-End Neural Speech Synthesizer

TL;DR

This paper presents an end-to-end neural speech synthesizer capable of generating emotional speech, addressing key issues like exposure bias and attention irregularity through novel model enhancements.

Contribution

The paper introduces improvements to Tacotron by using context vectors and residual connections to better handle emotional speech synthesis.

Findings

Successful training and speech generation for emotion labels

Improved attention alignment and stability in synthesis

Effective handling of exposure bias in neural speech models

Abstract

In this paper, we introduce an emotional speech synthesizer based on the recent end-to-end neural model, named Tacotron. Despite its benefits, we found that the original Tacotron suffers from the exposure bias problem and irregularity of the attention alignment. Later, we address the problem by utilization of context vector and residual connection at recurrent neural networks (RNNs). Our experiments showed that the model could successfully train and generate speech for given emotion labels.