Reproduce “Emilia2: A Large-Scale Dataset for Controllable Long-Form Speech Generation” 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
Qinke Ni, Dekun Chen, Junwen Qiu, Huan Liao, Xu Li, Xingwen Han, Wan Lin, Kunyu Feng and 9 more
Emilia2 builds a long-form speech dataset from continuous recordings, creating short utterances, single-speaker passages, and multi-speaker dialogue with transcripts, style instructions, and nonverbal-vocalization tags. Its long-form examples preserve pauses and transitions rather than stitching independent clips together. The 229,388-hour corpus spans 41 languages; a 0.6B-backbone model trained on it achieves the lowest Chinese character and English word error rates among compared systems on TTSD-Eval dialogue, not across every task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.