ASK-QA
收藏资源简介:
ASK-QA是由哥伦比亚大学和谷歌的研究团队创建的第一个多轮口语对话数据集,旨在解决多模态语音建模中的模糊用户请求和动态评估输入问题。该数据集包含5985条训练对话,涵盖了多种口音、语速和音调,且每条对话的音频长度较长,平均为1分41秒。数据集的创建过程涉及从文本对话数据集Abg-CoQA开始,通过文本到语音合成技术生成多轮对话的音频文件。ASK-QA主要应用于智能对话系统,特别是需要处理复杂口语交互的场景,如智能助手和客户支持系统。
ASK-QA is the first multi-turn spoken dialogue dataset developed by research teams from Columbia University and Google, designed to solve the problems of ambiguous user requests and dynamic input evaluation in multimodal speech modeling. The dataset includes 5985 training dialogue sessions, which cover a wide range of accents, speaking rates and pitches, with each dialogue having a relatively long audio duration, averaging 1 minute and 41 seconds. The dataset was created starting from the text-based dialogue dataset Abg-CoQA, and audio files for the multi-turn dialogues were generated using text-to-speech synthesis technology. ASK-QA is primarily applied to intelligent dialogue systems, particularly scenarios involving complex spoken interactions such as smart assistants and customer support systems.




