数据堂—1,136小时美式英语自然对话手机采集语音数据

1136小时美式英语自然对话手机采集语音数据由1000余名发音人参与录制，以自然方式进行交流，针对给定的数个话题自由发挥，领域广泛，语音自然流利，符合实际对话场景。1136小时美式英语自然对话手机采集语音数据由人工转写文本，准确率高

The 1136-hour American English natural conversational speech dataset was collected using mobile phones, with recordings sourced from over 1,000 speakers. Participants engaged in natural communication, speaking freely on multiple assigned topics across a wide range of domains. The collected speech features natural and fluent utterances that align with real-world conversational scenarios. Furthermore, the dataset is paired with high-accuracy manual transcriptions.

搜集汇总

背景与挑战

背景概述

该数据集包含1,136小时通过手机采集的美式英语自然对话语音，由超过1,000名说话者参与录制，涵盖多样话题，语音格式为16kHz单声道WAV，并配有高精度人工转录。它主要用于测试美式英语识别模型。