ASPIRE

ASPIRE数据集是由爱丁堡龙比亚大学等机构创建的，首个在真实嘈杂环境中记录的高质量视听双耳语音数据集，主要在餐厅和食堂等场景中录制。该数据集包含10000条语音，由5名年龄在23至55岁之间的英国本土演讲者录制，每位演讲者录制2000条语音，涵盖多种英语口音。数据集的创建过程包括在繁忙时段使用iPad mini2和Zoom H4n pro录音设备进行录制，确保语音和视频的高质量同步。ASPIRE数据集主要用于评估和开发新的视听语音增强技术，特别是在处理多模态语音增强系统中的背景噪声和语音混响问题。The ASPIRE dataset, developed by institutions including Edinburgh Napier University, is the first high-quality audiovisual binaural speech dataset recorded in real noisy environments, and was primarily collected in scenarios such as restaurants and canteens. It contains 10,000 speech utterances recorded by 5 native British speakers aged between 23 and 55, with each speaker contributing 2,000 utterances, covering a variety of English accents. The dataset was captured during peak hours using recording devices including iPad mini 2 and Zoom H4n Pro, ensuring high-quality synchronization of audio and video. The ASPIRE dataset is primarily intended for evaluating and developing novel audiovisual speech enhancement technologies, particularly for addressing background noise and speech reverberation in multimodal speech enhancement systems.

搜集汇总

背景与挑战

背景概述

ASPIRE数据集是首个在真实嘈杂环境（如餐厅和食堂）中录制的高质量视听双耳语音数据集，包含10000条语音，由5名英国本土演讲者录制，涵盖多种英语口音。它主要用于评估和开发视听语音增强技术，以处理背景噪声和语音混响问题。