AISHELL-4

AISHELL-4是由西北工业大学等机构合作创建的大型普通话语音数据集，专为会议场景设计。该数据集包含211个会议记录，总时长120小时，涵盖4至8名发言者，具有真实的声学特性和丰富的自然对话特征。数据集创建过程中，通过8通道圆形麦克风阵列收集，确保了高质量的语音记录和准确的转录。AISHELL-4主要用于推动多发言人语音处理的研究，包括语音前端处理、语音识别和发言人分割等，旨在解决实际会议场景中的语音技术挑战。

AISHELL-4 is a large-scale Mandarin speech dataset jointly created by Northwestern Polytechnical University and other institutions, specifically designed for conference scenarios. The dataset contains 211 meeting recordings with a total duration of 120 hours, involving 4 to 8 speakers, and features realistic acoustic properties and rich natural conversation characteristics. During the dataset construction, the recordings were collected using an 8-channel circular microphone array, ensuring high-quality speech records and accurate transcriptions. AISHELL-4 is primarily used to advance research on multi-speaker speech processing, including speech front-end processing, speech recognition and speaker diarization, aiming to address speech technology challenges in real-world conference scenarios.