Reproduce “VoxCanvas-41M: A Large-Scale Bilingual Dataset for Instruction-Based Speech Editing and Its Learning Dynamics” 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
Zhikang Niu, Wenming Tu, Yalin Li, Yanru Huo, Junxi Liu, Tianrui Wang, Ruiqi Yan, Junxuan Liang and 3 more
The authors introduce VoxCanvas, a pipeline and dataset for instruction-based speech editing, illustrated by “five Paris universities discovered” becoming “two London colleges found” in the released synthetic-to-real direction. VoxCanvas-41M contains about 41 million Chinese and English pairs. On English content editing in SpeechEditBench, VoxCanvas-Edit scores 96.78 ACC, versus 94.22 for AuK fine-tuned on GigaEdit-L; this comparison does not establish gains across all tasks or explain their cause.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.