Keynote Speaker

Professor Wu Zhizheng

Professor Wu Zhizheng is currently an Associate Professor at The Chinese University of Hong Kong, Shenzhen. He holds the position of Deputy Director at the Shenzhen Key Laboratory of CrossModal Cognitive Computing. Professor Wu has been consistently listed in Stanford University’s “World’s Top 2% Scientists” and has received multiple Best Paper Awards. He earned his Ph.D. from Nanyang Technological University and has held research and leadership roles at internationally renowned institutions, including Meta (formerly Facebook), Apple, the University of Edinburgh, and Microsoft Research Asia. Professor Wu has initiated several influential open-source projects, such as Merlin, Amphion, and Emilia, which have been adopted by over 700 organizations worldwide, including OpenAI. Notably, Amphion has topped GitHub’s trending list multiple times, while Emilia has become the most popular audio dataset (Most Liked) on HuggingFace. He also initiated and organized the first ASVspoof Challenge and the first Voice Conversion Challenge and served as the organizer of the Blizzard Challenge 2019, a prestigious international speech synthesis competition. Currently, Professor Wu serves on the editorial boards of IEEE/ACM Transactions on Audio, Speech and Language Processing and IEEE Signal Processing Letters, and is the General Chair of the IEEE Spoken Language Technology Workshop 2024.

Reinforcement Learning for text-to-speech synthesis

Reinforcement Learning (RL) is emerging as a powerful paradigm for refining Text-to-Speech (TTS) systems, moving beyond standard supervised training to fine-tune nuanced qualities like naturalness and expressiveness. While these methods improve general performance, they often fail to address specific, critical flaws. One of the most significant challenges for modern zero-shot TTS is maintaining intelligibility, especially for difficult inputs like tongue twisters, repeated words, cross-lingual prompts, or code-switching. This talk will explore how targeted preference alignment, a form of RL, can solve these issues. We present a case study based on the paper “Advancing Zero-shot Text-to-Speech Intelligibility across Diverse Domains via Preference Alignment.” The approach introduces a new Intelligibility Preference Speech Dataset (INTP) and extends the Direct Preference Optimization (DPO) framework for TTS. Alignment with INTP significantly boosts intelligibility and enhances overall speech quality. We also demonstrate weak-to-strong generalization, improving even state-of-the-art models, and show the potential for iterative refinement. This work highlights a strategic shift from general enhancement to precisely correcting flaws, charting a path toward truly robust and reliable TTS.