数据堂—234小时日语手机采集语音数据

234小时日语手机采集语音数据采集自799名日本本土发音人，在安静的室内、街道、餐馆等环境下的录音数据。录音内容广泛，包括21万句常用日语书面语及常用日语口语。234小时日语手机采集语音数据的文本转写句错误率小于5%。匹配主流安卓、苹果系统手机

This 234-hour Japanese speech dataset was collected via mobile phones from 799 native Japanese speakers. Recordings were obtained in various real-world environments including quiet indoor spaces, streets, restaurants and other daily scenarios. The dataset's transcriptions cover 210,000 sentences of standard daily written and spoken Japanese. The sentence-level transcription error rate of this dataset is less than 5%. It is compatible with mainstream Android and Apple iOS mobile devices.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于日语语音识别模型测试的语音数据集合，总时长为234小时，通过手机采集自799名日语母语者，涵盖安静室内等多种环境。数据以16kHz WAV格式提供，版权归数据堂所有，可供商业使用。