Learning Latent Representations for Style Control and Transfer in End-to-end Speech Synthesis

TLDR

The study introduces a Variational Autoencoder into an end‑to‑end speech synthesis model to learn unsupervised latent representations of speaking styles. The model uses a VAE whose recognition network infers style embeddings that are fed into the TTS network for synthesis, while training incorporates techniques to prevent KL divergence collapse. The learned style embeddings exhibit disentanglement, scaling, and combination, enabling effective style control and outperforming the GST model in ABX preference tests.

Abstract

In this paper, we introduce the Variational Autoencoder (VAE) to an end-to-end speech synthesis model, to learn the latent representation of speaking styles in an unsupervised manner. The style representation learned through VAE shows good properties such as disentangling, scaling, and combination, which makes it easy for style control. Style transfer can be achieved in this framework by first inferring style representation through the recognition network of VAE, then feeding it into TTS network to guide the style in synthesizing speech. To avoid Kullback-Leibler (KL) divergence collapse in training, several techniques are adopted. Finally, the proposed model shows good performance of style control and outperforms Global Style Token (GST) model in ABX preference tests on style transfer.