Expressive Speech Synthesis via Modeling Expressions with Variational Autoencoder

• DOI: 10.21437/interspeech.2018-1113
• Publication Date: 2018-08-28T09:55:42Z
• Authors: Kei Akuzawa; Yusuke Iwasawa; Yutaka Matsuo
• Keywords: Autoencoder

Recent advances in neural autoregressive models have improve the performance of speech synthesis (SS).However, as they lack ability to model global characteristics (such speaker individualities or speaking styles), particularly when these not been labeled, making SS systems more expressive is still an open issue.In this paper, we propose combine VoiceLoop, model, with Variational Autoencoder (VAE).This approach, unlike traditional systems, uses VAE explicitly, enabling expressiveness synthesized be controlled unsupervised manner.Experiments using VCTK and Bliz-zard2012 datasets show helps VoiceLoop generate higher quality control expressions its by incorporating into generating process.