<html><body><div style="font-family: arial, helvetica, sans-serif; font-size: 12pt; color: #000000"><div>During commuting and resting, audiobook APPs like 喜马拉雅 has become a popular choice for people to spend their eye-free times. However, the production of audiobooks are very time-consuming and also requires professional skills and devices. TTS is a promising alternative for audiobook production. However, traditional TTS systems are lacking in expressive features like emotions and the distinction of different characters, which may make the user boring. <br><br>In this ICASSP 2021 paper: "<strong>A Chapter-Wise Understanding System for Text-To-Speech in Chinese Novels</strong>", researchers from ByteDance AI Lab propose to use a "novel" NLU system to improve the expressiveness of TTS. The resulting system has been applied in 番茄免费小说 and 番茄畅听, and achieved remarkable success (over a billion downloads in HUAWEI APP store). <br><br>As one of the contributors of the project, I will introduce the interesting points inside and beyond what's written in the paper. Hope you can draw inspirations from it.<br><br>Related:<br>My previous talk on TTS, covering the basics of a TTS system (mainly frontend)<br><a data-mce-href="https://tx9k8kh0gj.feishu.cn/file/boxcn8s0uCVXmd1AYbW8HIMDhHb" href="https://tx9k8kh0gj.feishu.cn/file/boxcn8s0uCVXmd1AYbW8HIMDhHb">[2019-12-11][zhiling] TTS frontend:NLP as backbone of speech synthesis</a><br><br>Time: Wed 4:00pm<br><br>Venue: SEIEE <strong>3-526A</strong><br><br>Best,<br>Zhiling<br></div></div></body></html>