Projects

I try to build artifacts that are useful beyond a single paper: datasets, demos, live systems, and creative tools.

Featured

LEMAS
150K+ hour, 10-language multilingual speech corpus with word-level timestamps, plus LEMAS-TTS and LEMAS-Edit benchmark models.
SVLive
Browser-based live coding music playground combining Strudel, Vital, and LLM assistance for real-time editable synthesis.
VoicePlatform
Unified audio/video AI workflow integrating enhancement, ASR, TTS/VC, speech editing, video editing, and translation.
Robot Speech Interaction
Omni speech-interaction LLM post-training and evaluation for embodied robotics and real-time human-robot interaction.

Open Source / Data

OpenVideo
672+ hour / 106K+ clip / 720p text-to-video dataset with annotation, validation, and migration tooling.
Multi-speaker VC
Expressive multi-speaker voice conversion demo and publication page for the ACM ICMI 2021 work.