数据堂—2,657小时普通话自然对话手机采集语音数据

2,657小时普通话自然对话手机采集语音数据由4491名发音人参与录制，以自然方式进行面对面交流，不限制话题自由发挥，领域广泛，语音自然流利，符合实际对话场景。2,657小时普通话自然对话手机采集语音数据由人工转写文本，准确率高。

This 2,657-hour Mandarin natural conversational speech dataset was collected via mobile phones, with 4,491 speakers participating in the recording. The recordings were conducted in natural face-to-face conversations, with no topic restrictions to allow free topic development, covering a wide range of domains. The speech is natural and fluent, aligning with real-life conversational scenarios. Additionally, all the speech data was manually transcribed into text with high accuracy.

搜集汇总

背景与挑战

背景概述

该数据集包含2,657小时的普通话自然对话语音，由4,491名发音人录制，语音自然流利，适用于中文语音识别模型的测试任务。数据格式为16kHz的wav文件，人工转写文本准确率高，标注详细，适合商用。