数据堂—513小时日语语音数据_对话（电话）

513小时日语自然对话电话信道语音数据由800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。人工转写文本，准确率高

This dataset contains 513 hours of Japanese natural conversational telephone-channel speech, recorded by more than 800 speakers. Participants conducted natural, spontaneous exchanges centered on several pre-specified topics across diverse domains. The recorded speeches are natural and fluent, fully consistent with real-world conversational scenarios. All utterances have been manually transcribed with high accuracy.

搜集汇总

背景与挑战

背景概述

该数据集包含513小时日语电话对话语音，由878名日本母语者录制，内容涵盖多个话题的自然交流，语音质量清晰且标注准确。它主要用于语音识别和说话人验证任务，句子准确率达到95%。