M3-SLU/M3-SLU-sample

M3-SLU-sample This repository provides 100-sample subsets of the M3-SLU benchmark for two tasks: Task 1: Speaker-Attributed Question Answering Task 2: Speaker Attribution Matching (True/False) Each example includes multi-speaker conversational audio, instruction/question/answer pairs, transcripts, and metadata (n_speakers, data_source).The full dataset (10K+ samples) will be released later in 2025.