A Cocktail-Party Benchmark: Multi-Modal dataset and Comparative Evaluation Results

TL;DR

The paper introduces MCoRec, a multi-modal benchmark for context-aware recognition in cocktail-party conversations, addressing overlapping speech with audio, visual, and contextual cues. It presents a dataset of up to eight participants across up to four simultaneous conversations, recorded with a central 360° view and participant-specific video and audio, plus annotation and synchronization procedures. Baselines cover Active Speaker Detection, AVSR (including a finetuned AV-HuBERT model), and conversation clustering, with a novel Joint ASR-Clustering metric that combines transcription accuracy and clustering quality. The work provides a reproducible benchmark and baseline results, highlighting the benefits of multi-modal data while identifying substantial room for improvement in both transcription and clustering for natural multi-party interaction.

Abstract

We introduce the task of Multi-Modal Context-Aware Recognition (MCoRec) in the ninth CHiME Challenge, which addresses the cocktail-party problem of overlapping conversations in a single-room setting using audio, visual, and contextual cues. MCoRec captures natural multi-party conversations where the recordings focus on unscripted, casual group chats, leading to extreme speech overlap of up to 100% and highly fragmented conversational turns. The task requires systems to answer the question "Who speaks when, what, and with whom?" by jointly transcribing each speaker's speech and clustering them into their respective conversations from audio-visual recordings. Audio-only baselines exceed 100% word error rate, whereas incorporating visual cues yields substantial 50% improvements, highlighting the importance of multi-modality. In this manuscript, we present the motivation behind the task, outline the data collection process, and report the baseline systems developed for the MCoRec.

Figures (1)

• Figure 1: Recording setup of a session in the MCoRec dataset.