Reproduce “LiveSynth: A Streaming Neural Synthesizer for Instrument Cloning and Text-to-Instrument” 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
Kyungsu Kim, Yejin Kim, Kyogu Lee
LiveSynth streams audio conditioned on a reference recording or text prompt, responding frame by frame to incoming MIDI. It generates continuous audio latents and decodes them causally, with measured processing fitting each 10 ms frame. On 530 held-out NSynth clips, it scores 0.540 versus TokenSynth’s 0.465 on MERT similarity and 13.1 versus 26.0 on MSS loss, but scores lower on note adherence and production quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.