Summary of the NOTSOFAR-1 challenge: Highlights and learnings

Open Webpage

Igor Abramovski, Alon Vinnikov, Shalev Shaer, Naoyuki Kanda, Xiaofei Wang, Amir Ivry, Eyal Krupka

Published: 01 Jan 2025, Last Modified: 19 Sept 2025Comput. Speech Lang. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Comprehensive review of the results of CHiME-8 NOTSOFAR-1 Challenge.•Fine-tuning DASR pipelines with real meeting data improved their accuracy.•So did ASR adaptation to far-field audio preprocessed by an acoustic front-end.•GSS dominates multi-channel speech separation; neural alternatives are emerging.•Geometry-agnostic systems achieve accuracy comparable to known-geometry ones.