Reproduce “STAM-ASR: Speaker-Temporal Anchoring with Memory for Multi-Speaker ASR” 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
Victor Tolulope Olufemi, Syeda Faiza Ahmed Sara, Shammur Absar Chowdhury
STAM-ASR transcribes multi-speaker meetings by conditioning an AudioLLM’s transcript representation on speaker and timing cues from its own audio features. It carries context in separate speaker-turn and chronological conversation memories, without waveform separation or an external diarization frontend. With predicted speaker activity, adding both memories lowers ICSI cpWER from 51.8% to 47.7%, but raises LibriCSS cpWER from 51.1% to 53.3%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.