Predicting Turn-Taking and Backchannel in Human-Machine Conversations Using Linguistic, Acoustic, and Visual Signals

2025-05-19

AI Summary

To address the scarcity of high-quality multimodal annotated data for turn-taking and backchannel prediction in human–computer dialogue, this paper introduces MM-F2F—the first large-scale, fine-grained multimodal face-to-face dialogue dataset (210 hours of video, 1.5M words, 20M frames), featuring frame-level annotations for both turn-taking and backchannel behaviors. We propose an end-to-end multimodal dynamic fusion framework comprising: (i) a novel automated multimodal acquisition pipeline; (ii) a modality-strongly-coupled modeling mechanism enabling flexible input from any subset of text, speech, and video modalities; and (iii) integrated self-supervised pretraining, cross-modal alignment, and temporal probabilistic prediction. Experiments demonstrate state-of-the-art performance, with absolute F1-score improvements of +10% for turn-taking prediction and +33% for backchannel prediction. The MM-F2F dataset and source code are publicly released.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

This paper addresses the gap in predicting turn-taking and backchannel actions in human-machine conversations using multi-modal signals (linguistic, acoustic, and visual). To overcome the limitation of existing datasets, we propose an automatic data collection pipeline that allows us to collect and annotate over 210 hours of human conversation videos. From this, we construct a Multi-Modal Face-to-Face (MM-F2F) human conversation dataset, including over 1.5M words and corresponding turn-taking and backchannel annotations from approximately 20M frames. Additionally, we present an end-to-end framework that predicts the probability of turn-taking and backchannel actions from multi-modal signals. The proposed model emphasizes the interrelation between modalities and supports any combination of text, audio, and video inputs, making it adaptable to a variety of realistic scenarios. Our experiments show that our approach achieves state-of-the-art performance on turn-taking and backchannel prediction tasks, achieving a 10% increase in F1-score on turn-taking and a 33% increase on backchannel prediction. Our dataset and code are publicly available online to ease of subsequent research.

Problem

• Predicting turn-taking and backchannel in human-machine conversations
• Overcoming dataset limitations with multi-modal signal collection
• Developing an adaptable end-to-end framework for conversation analysis

Innovation

• Automatic data collection pipeline for conversation videos
• Multi-Modal Face-to-Face dataset with diverse annotations
• End-to-end framework for multi-modal turn-taking prediction