自然人声纹语音识别训练库

       本数据库采用手机平台录制，分别有三种不同系统的手机设备收录发音人的声音：安卓系统手机、苹果手机与Windows系统手机。        发音人按照设计好的文本录音，文本所涉及内容基本来自日常用语、新闻、网上聊天等渠道。        在录制过程中，为研究发音人其声音隔一段时间是否会产生变化，每位发音人的录音分为两次完成，两次间隔不少于一周的时间，并且录音内容互不重复。每次录音时，发音人需要在30分钟内以自然放松的语气和语速，录制单句、数字串、电话号码、命令词、长段落等106句语料。        经人工校对、筛选过滤和质检后，该数据库保留了19万句有效语料，所有语料都由母语发音人做了转写和标注，整体准确率不低于95%。此外，该数据库提供一个中文普通话发音词典。King-ASR-620

This database was recorded using mobile phone platforms, with voice data from speakers collected via three types of mobile devices with different operating systems: Android smartphones, iPhones, and Windows smartphones.

Speakers recorded speech based on pre-designed scripts, the content of which mainly originates from daily expressions, news articles, online chats and other sources.

During the recording process, to investigate whether speakers' voices change over time, each speaker completed two separate recording sessions, with an interval of no less than one week between the two sessions, and the recorded content of the two sessions was completely non-overlapping. For each session, speakers were required to record 106 speech samples including single sentences, digit sequences, phone numbers, command terms, long paragraphs and other types within 30 minutes, using a natural and relaxed tone and speech rate.

After manual proofreading, filtering and quality inspection, this database retains 190,000 valid speech samples. All samples were transcribed and annotated by native Mandarin speakers, with an overall accuracy rate of no less than 95%. In addition, this database provides a Standard Mandarin Chinese pronunciation dictionary. King-ASR-620