AISHELL-DMASH 中文普通话麦克风阵列家居场景语音数据库

AISHELL-DMASH 数据集是在两个不同房间的真实智能家居场景中记录的，该数据集包含 30000 小时的语音数据。录音设备包括一个近距离麦克风和位于房间 7 个不同位置的 7 组设备。一组录音设备包括一部 iPhone 、一部 Android 手机、一部 iPad 、一个麦克风以及一个半径为 5cm 的圆形麦克风阵列。该数据集包含 511 位说话者，每个说话者访问 3 次，间隔 7-15 天。 AISHELL-DMASH 数据集由专业语音标注人员转录，单词准确率达 98%，可用于声纹识别、语音识别、唤醒词识别等研究。

The AISHELL-DMASH dataset was recorded in real smart home scenarios across two distinct rooms, containing 30,000 hours of speech data. The recording setup includes one close-talking microphone and seven sets of devices deployed at seven different locations within each room. Each set of recording devices consists of an iPhone, an Android smartphone, an iPad, a standalone microphone, and a circular microphone array with a radius of 5 cm. The dataset comprises 511 unique speakers, each of whom visited the recording scenarios three times, with an interval of 7 to 15 days between consecutive visits. The AISHELL-DMASH dataset was transcribed by professional speech annotators, achieving a word accuracy rate of 98%, and is suitable for research in voiceprint recognition, speech recognition, wake word recognition, and other related fields.

搜集汇总

背景与挑战

背景概述

AISHELL-DMASH是一个包含30000小时中文普通话语音数据的家居场景数据集，由511位说话者在两个不同房间的真实智能家居环境中录制，使用多种录音设备（包括近距离麦克风和7组不同位置的设备）。该数据集经过专业转录，单词准确率达98%，适用于声纹识别、语音识别和唤醒词识别等研究。