Experience

Co-founded the company and helped design and build video direct translation and comic-drama dubbing products from zero to one. My work covers product definition, user and client scenario analysis, data and model pipelines, prototype iteration, delivery feedback loops, and speech generation quality evaluation, turning low-resource TTS and dubbing research into practical ToB/C product capabilities. Company website: luoji.cn.

Worked in the Foundation Model team on Instruct TTS based on CosyVoice and LLASA, and on continuous-feature-based speech synthesis and controllable singing voice synthesis based on VibeVoice and Ming-Omni-TTS for in-car voice interaction. The work covered data construction, model training, evaluation, and product-oriented validation for controllable, high-quality speech generation.