SynthWOZ

该数据集名为SynthWOZ，是一个全面的多轮合成音频数据集，旨在辅助基于音频的对话状态跟踪任务。它涵盖了七个不同领域的对话内容，同时兼具合成音频和真实的人类录音。该数据集中包含了十个不同的合成语音，这些语音在性别和口音上各不相同，并且还包含了来自十位发言人的真实录音，以评估模型在实际环境中的表现。这一任务专注于基于音频的对话状态跟踪（Dst）。

The dataset named SynthWOZ is a comprehensive multi-turn synthetic audio dataset intended to support research on audio-based dialogue state tracking tasks. It encompasses conversational content across seven distinct domains, and contains both synthetic audio and authentic human speech recordings. Specifically, the dataset includes ten distinct synthetic voices with varying genders and accents, as well as real speech recordings from ten speakers to enable evaluation of model performance in real-world scenarios. This task centers on audio-based dialogue state tracking (DST).