Reproduce “Spatially-Guided Speaker Extraction Based on Relative Position Cues” 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
Aviad Eisenberg, Shlomo E. Chazan, Sharon Gannot
The authors introduce relative position cues (RPC) for extracting a chosen voice from overlapping speech captured by a microphone array. The cue tells the model whether the target is nearer, farther, left, or right relative to another speaker, and conditions a multichannel model. In a four-microphone U-Net, RPC reaches 11.00 dB SI-SDR versus 10.29 dB with absolute DOA on a test set with small distance and large angular separation. The evaluation assumes concurrent speakers and a valid relation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.