MultiDialog

该数据集名为MultiDialog，是一个大型的多模态口语对话语料库，包含了大约340小时的对话录音，这些录音来自于约9000个对话，对话双方基于开源领域对话数据集TopicalChat进行交流，并采用了音视频语音记录。此外，该数据集还包含了情感标注，涉及八种情感的平行音视频记录和情感标注，旨在促进多模态合成和对对话系统的研究。该数据集的规模大约为9000个对话，其任务重点在于多模态对话的生成和合成。

This dataset is named MultiDialog, a large-scale multimodal spoken dialogue corpus containing approximately 340 hours of dialogue recordings sourced from around 9,000 dialogues. The conversational participants communicate based on the open-domain dialogue dataset TopicalChat, and all recordings are in audio-visual format. Additionally, this dataset includes parallel audio-visual recordings and emotion annotations covering eight distinct emotions, aiming to promote research on multimodal synthesis and dialogue systems. With a scale of roughly 9,000 dialogues, its core tasks focus on the generation and synthesis of multimodal dialogues.

搜集汇总

背景与挑战

背景概述

MultiDialog是一个英语多模态对话数据集，包含8,733个对话和187,859个话语，每个话语平均长度为6.51秒。数据集提供了音频文件、转录文本和情感标注，支持多模态对话生成、自动语音识别和文本到语音等任务。