Semi-supervised Training for Improving Data Efficiency in End-to-end Speech Synthesis

Although end-to-end text-to-speech (TTS) models such as Tacotron have shown excellent results, they typically require a sizable set of high-quality pairs for training, which are expensive to collect. In this paper, we propose a semi-supervised training framework to improve the data efficiency of Ta...

Full description

Saved in:
Bibliographic Details
Main Authors: Chung, Yu-An, Wang, Yuxuan, Hsu, Wei-Ning, Zhang, Yu, Skerry-Ryan, R.J.
Format: Conference Proceeding
Language:English
Subjects:
Acoustics
Artificial neural networks
Context modeling
data efficiency
Data models
Decoding
pre-training
semi-supervised learning
Tacotron
text-to-speech
Training
Training data
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Although end-to-end text-to-speech (TTS) models such as Tacotron have shown excellent results, they typically require a sizable set of high-quality pairs for training, which are expensive to collect. In this paper, we propose a semi-supervised training framework to improve the data efficiency of Tacotron. The idea is to allow Tacotron to utilize textual and acoustic knowledge contained in large, publicly-available text and speech corpora. Importantly, these external data are unpaired and potentially noisy. Specifically, first we embed each word in the input text into word vectors and condition the Tacotron encoder on them. We then use an unpaired speech corpus to pre-train the Tacotron decoder in the acoustic domain. Finally, we fine-tune the model using available paired data. We demonstrate that the proposed framework enables Tacotron to generate intelligible speech using less than half an hour of paired training data.
ISSN:2379-190X
DOI:10.1109/ICASSP.2019.8683862