0x3/CABankSakuraCHJP

该数据集是CallHome语料库的日语部分，专注于电话通话语音收集。数据收集由LDC组织，通过互联网、出版物和个人联系招募了200名呼叫发起者。参与者使用LDC维护的免费机器人操作员拨打电话，通话过程被录音，且双方均知情同意。每次通话最长30分钟，完成后呼叫者获得20美元报酬。语料库包含200个通话的转录，其中80个用于训练，20个用于开发测试，100个用于评估测试（本次发布仅包含20个评估测试通话）。训练和开发测试通话转录了连续的10分钟区域，评估测试通话转录了5分钟区域。数据收集后进行了人工审核，以确保语言正确、录音质量，并描述转录区域的特征，如信道质量、说话者数量、性别等。数据集旨在支持语音识别和说话人识别等任务。

This is the Japanese portion of the CallHome corpus, focusing on telephone call speech collection. The data collection was organized by LDC, recruiting 200 call originators via the internet, publications, and personal contacts. Participants made calls using a toll-free robot operator maintained by LDC, with recordings taken after informed consent from both parties. Each call lasted up to 30 minutes, and callers were paid $20 upon completion. The corpus includes transcriptions for 200 calls, with 80 designated for training, 20 for development test, and 100 for evaluation test (only 20 evaluation test calls are released in this publication). For training and development test calls, a contiguous 10-minute region was transcribed, while for evaluation test calls, a 5-minute region was transcribed. After collection, human audits were conducted to verify language usage, recording quality, and to describe transcribed regions, including channel quality, number of speakers, gender, and other attributes. The dataset is intended for tasks such as speech recognition and speaker identification.