Reproduce “Fish Audio Drama 3 Technical Report” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dawei Han, Shidong Li, Shijia Liao, Jiahua Liu, Songting Liu, Qingzheng Wang, Yuxuan Wang, Xin Chen and 8 more
Drama 3 can revise speech through conversation, keeping prior takes, voice references, and directions in context. It also edits bounded spans by infilling them from surrounding audio. On Step-Audio-Edit, its judged style accuracy reaches 81.8% at revision iteration 3, versus 63.8% for Step-Audio-EditX in the reported Prompt-Fixed comparison. Its emotion accuracy remains lower than Step-Audio-EditX at iteration 3; scores come from separate evaluation runs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.