Reproduce “SciPhi-Chat: Training a Spatial Audio Agent for Question Answering, Dialogue, and Tool Use” 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

Gijs Wijngaard, Hannes Gamper, Sebastian Braun

SciPhi-Chat answers spatial-audio questions by building a structured scene description from four-channel first-order ambisonics and retaining it with earlier answers across turns. It can request signal-processing measurements from tools. On MASU, it scores 44.9% on the single-turn track, versus 37.5% for a scene-description-plus-reasoner baseline; the score averages nine task-group scores and does not isolate which system component accounts for the difference.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.