数据堂—1,535小时中英混读手机采集语音数据

1,535小时中英混读手机采集语音数据由3972名中国本土人员参与录制，口音覆盖七大方言区。录音文本均为中英混合句子，涵盖通用场景及人机交互场景，内容丰富，转写精准。可用于改善语音识别系统对中英混读语音的识别效果

This dataset consists of 1,535 hours of mobile-collected code-switching Mandarin-English speech data, recorded by 3,972 native Chinese individuals. The speech samples cover accents from seven major Chinese dialect regions. All accompanying transcripts are mixed Mandarin-English sentences, encompassing both general daily scenarios and human-computer interaction (HCI) scenarios, featuring rich content and precise transcriptions. This dataset can be utilized to improve the speech recognition performance of systems designed for code-switching Mandarin-English speech.

搜集汇总

背景与挑战

背景概述

该数据集包含1,535小时通过手机采集的中英混读语音数据，采样率为16kHz，由3,972名中国母语者录制，覆盖七大方言区。它用于测试中英混读语音识别模型，录音环境安静且无回声，数据以16位未压缩WAV单声道格式存储。