耳如
隐私政策
1. 我们是谁
耳如(英文名 Earu,网址 https://earu.ai)是一个练听力的视频播放器,帮你用自己爱看的视频学第二语言,有网页版和桌面版(包括从 Mac App Store 安装的耳如 Mac 版)。它由一位个人开发者运营(下称「耳如运营者」或「我们」),没有成立公司,运营者在美国。
本政策说明你使用耳如网站和服务时,我们收集哪些信息、怎么用、存在哪里、保存多久,以及你可以怎样查询和删除。
2. 要点
- 你的视频文件只在你自己的电脑或手机上播放,不会上传到我们的服务器。
- 为了翻译,耳如会把视频的原声字幕文字和片名发到我们的服务器,由服务器交给 AI 服务商 翻译成你的母语(默认是界面语言,可以在学习设置里改)。字幕和译文按内容全站共用(见 3.5)。 在耳如 Mac 版里,登录前要先同意交给 AI 服务商处理,不同意没法登录(见 7)。
- 跟读时,每读一遍的录音都会经我们的服务器发给 AI 服务商 识别成文字;服务器不保存录音,只保存识别出的文字、分数和没读准的词。用来回放的录音和排对时剪下的几秒原声只存在你这台设备的浏览器里(见 3.8、3.9)。 在耳如 Mac 版里,登录前要先同意交给 AI 服务商处理,不同意没法登录(见 7)。
- 你可以用手机号加短信验证码登录,也可以用 Google 账号登录(桌面版)。我们只发验证码短信,不发营销短信。
- 我们不出售、不出租你的个人信息,也不会把它提供给任何第三方用于营销。
- 手机号账号:在安卓手机或平板上,以及从 Mac App Store 安装的耳如 Mac 版里,你可以随时在设置里自己删除账号和全部数据,用短信验证码确认。只用 Google 登录的账号:在桌面版里(Windows 版,或从 earu.ai 下载的 Mac 版)同样可以在设置里自己删除,通过重新登录 Google 确认。步骤见《删除账号》;也可以发邮件到 support@earu.ai 让我们删除。
3. 我们收集哪些信息
3.1 登录信息(手机号、Google 账号)
- 手机号:你登录时填写的手机号。世界各国和地区的手机号都可以;免费电话、收费号码、卫星电话这类收不了短信的号码不支持。用手机号登录的账号,手机号就是你的账号。
- Google 账号信息:只在你用 Google 账号登录,或在设置里绑定 Google 账号时才有。我们从 Google 收到并保存三样东西:Google 账号的唯一标识、邮箱地址和显示名称,只用来认出你的账号、让你登录。我们拿不到也不保存你的 Google 密码,也不向 Google 申请读取邮件、云端硬盘等其他权限。登录在耳如应用内打开的 Google 官方页面上完成,适用 Google 自己的隐私政策。已有手机号账号可以在设置里绑定或解除绑定 Google 账号,绑定只是多一种登录方式。
- 账号 ID:服务器为每个账号生成的随机编号,学习记录按这个编号保存。
- 账号时间:账号创建时间、最近一次登录时间。
- 时区:第一次登录时从浏览器读取,用来决定你的「今天」从什么时候开始;可以在学习设置里修改。
- 登录令牌:登录成功后发给浏览器的一串随机字符,用来保持登录。浏览器每次请求都会带上令牌,服务器据此认出你。服务器上保存令牌和它对应的账号(手机号或 Google 账号)、创建时间、最近使用时间,以及这次登录是否经过短信验证。
- 验证码发送记录:每次获取验证码时,我们记下手机号、请求来源的 IP 地址、时间和结果;输错验证码也会记一笔(这一笔不记 IP)。这些记录只用来防止滥用、限制发送次数。
- 验证码:一次性使用,10 分钟内有效。中国大陆号码的验证码由我们的服务器生成,只保存加密钥的哈希、不保存验证码本身,2 天后删除;其他号码的验证码由 Twilio 的验证服务生成和核对,我们不保存。验证码不写进日志。
- 短信同意记录:每次给你发出验证码时,记下手机号、用途、同意说明的版本、发码的时间和次数、截短后的 IP 网段(不存完整 IP),用来证明你同意过接收验证码短信。
- 登录记录:每次登录成功时,记下时间、IP 地址,以及从浏览器自报的信息看出来的设备类型、操作系统和浏览器;每个账号只留最近 50 次,用来保障账号安全、排查问题。
- 推广来源:如果你是通过带来源标记的推广链接(例如 earu.ai/?src=xhs)第一次登录的,账号上会记下这个来源,用来了解各推广渠道带来了多少用户。
3.2 学习记录
在「排序」练习中每计一句分,服务器保存一条记录:时间和日期、字幕指纹和句子序号、句子原文、片名、片段起止时间、句子难度等级、得分、使用提示的次数、用时、循环次数等。此外还保存:每天的目标分值与达标情况(例如达标的时间);每天的打卡记录(打卡的时间、是怎么打上的——做完每日小测,或当天只要求排序时排序达标——以及当天哪几项是系统跳过的);「最近在播」——你最近打开过哪份字幕、在什么时间,每人每份字幕一条,用来校验计分,也用来统计每份字幕有多少人打开过(运营者只看到人数)。
在「跟读」练习中,服务器保存:每天抽到的句子(从你排对过的句子里挑,含句子原文、字幕指纹和序号、片名);每读一遍的识别文字(最多 200 字)、分数、没读准的词、合格与否,和这一遍录音的时长、大小、音量(用来判断是不是没录到声音),以及这一遍的录音格式、录音窗口(这一遍最长能录多久)、识别用了多久、这台设备上有没有这句的原声、出错时的说明(例如识别超时);以及换句、系统跳过的记录。录音本身不保存(见 3.9)。在「每日小测」中,服务器保存:每天出的词和题目、每道题的作答(选的哪一项、对错、是否超时、用时、听了几遍)、每个词的熟练程度和下次复习的日期;题目里带的整句记有出处(字幕指纹、第几句、片名)。学习计划里的「预计用时」,是用你最近的这些记录里的用时(排序从第一句到达标、跟读和每日小测从开始到做完)算出来的。
3.3 学习设置
每日分值(50 到 200 分)、循环时间、字幕显示偏好、难度筛选、难度档(入门 / 基础 / 进阶 / 挑战)、跟读的设置(合格线、连着几次没过就给「换一句」、录音时长)、每日小测的设置(答题倒计时)、时区。每日分值按「从哪天起是多少」保存,历次修改都会留着,用来正确判断过去每一天有没有达标、那天该做多少。跟读每天几句、每日小测每天几题不单独保存,由当天的每日分值算出来。
3.4 生词本和查词记录
- 生词本:你亲手加入的单词(原形和你点的词形)、加入日期、出处(字幕指纹、第几句、句子原文、片名、时间点)。把单词移出生词本时只做「已移出」标记,这一行仍会保留,以便以后认出「以前加过的词」。
- 查词记录:登录状态下查过的单词都会记录(看视频时查的,以及在生词本页面点词查的),包括查询时间、在哪个页面查的和出处(字幕指纹、第几句)。没登录时查词不记录。查词记录保留 90 天。生词本页面能看到看视频时查过的词(按单词汇总,显示次数和最近一次查的日期);那里的「一键清空」会删掉你的全部查词记录。
3.5 字幕文字、片名和时长
打开视频时,网页在你的电脑或手机上读取视频里内嵌的字幕,先把按字幕内容算出的「指纹」发给服务器,问有没有现成的译文;服务器上还没有这份字幕时,再发送:整理成句子的字幕文字、每句的时长和片名。视频本身不会发送。
片名是网页从文件名(或视频文件里写的标题)整理出来的,例如剧名、年份、季、集、集名;认不出这些时,片名就是去掉扩展名、方括号和画质标记之后的文件名。如果文件名里有你不想让别人看到的内容,请先改名再打开。
视频网站上的视频:在桌面版里看哔哩哔哩(B 站)或 YouTube 的视频时,登录状态下耳如会把这部视频的站点和视频号发给我们,查它是不是片库里的视频:片库里的视频,字幕直接从我们的服务器取,不用上传;片库以外的视频,耳如读取视频网站提供的字幕文字,同样先发指纹,需要时再发整理成句子的字幕文字、每句时长和视频页上的标题。这些视频的画面和声音我们不下载、不保存。
网盘和 NAS 里的视频:从网盘或 NAS 读一个视频的内嵌字幕可能很慢,所以登录状态下,网页读完一条字幕轨后,会把这条轨的原样内容(每句的起止时间和原始文字,可能带字幕的样式标记)连同文件指纹(按文件大小和文件开头一小段内容算出的哈希,不含文件名和路径)、字幕轨信息(轨号、编码、语言、轨名)和片长发到服务器,全站共用:别人打开同一个文件时,网页先从自己的文件里抽查几段,对得上才用。服务器记下哪个账号传过或确认过哪一版(每个账号最多留 5000 条),删除账号时一起删除。
服务器按指纹保存字幕原文、片名、每句时长、AI 生成的译文、注释、术语表和每句难度。同一份字幕全站只翻译一次、共用一份:其他用户打开内容相同的字幕时,会直接看到已有的译文,也可能在生词本的「出处」里看到第一位上传者那里整理出的片名。这份缓存里记有第一个上传它的账号的手机号(用于用量统计和费用控制),只有运营者能看到,不会展示给其他用户。我们还记录每份字幕被打开的次数和最后打开时间(只记数字;用来判断「同一个人同一天只算一次」的标记约 2 天后删除)。
字幕时间校准:有的视频字幕和声音对不上(往往越往后差得越多),播放器会自动校准你在学的语言那条字幕的时间。先在你的设备上算:从这条音轨里找出哪里有人在说话,和字幕的时间对一对。这一步不上传任何声音;算出来的结果(几个数字)连同字幕指纹、视频的时长、文件大小和音轨编码存在你的账号下,只给你自己用。只有设备上判断不出来时,网页才从这条音轨里截几段短的声音(一般 5 段、每段约 1 分钟)发到我们的服务器,由服务器交给 AI 服务商 的语音识别转成带时间的文字,算出字幕该挪多少;这些声音片段处理完就丢掉,不保存,也不写进日志。这样算出的结果按字幕指纹和视频的时长、大小全站共用,里面没有你的账号信息。 在耳如 Mac 版里,登录前要先同意交给 AI 服务商处理,不同意没法登录(见 7)。
片库视频的字幕对齐检测:在桌面版里看耳如片库里的 B 站 / YouTube 视频时,耳如在你的设备上找出声音里有人说话的位置,和字幕时间对一对,看字幕有没有整体偏早或偏晚。这一步不上传任何声音。登录状态下,如果有把握发现偏差在 0.5 秒以上,会把偏差的秒数、把握程度、偏差出现在视频的大致哪几段、检测方法的名称、是哪部视频、哪一版字幕和一个随机的安装编号发给我们,用来修正片库字幕;同一台设备上同一部视频的同一版字幕只报一次。报告记录里没有你的账号和 IP(账号只用来数每天报了几次,这个计数 2 天后删除;服务器的访问日志照常记录,见 3.7),安装编号只存加密钥后的哈希;报告在最近一次上报 1 年后删除。
3.6 使用量统计
按账号记录每天的使用量:调用 AI 的次数和用量(token 数)、翻译的句数、直接从缓存拿到的句数、新上传的字幕数等,以及最近一次使用时间。用来控制成本、执行每日上限。
使用时长和在线状态:登录状态下,耳如会不时告诉服务器你是不是正在使用(窗口在前台并且最近有操作,或者视频正在播放)。服务器按账号记下每天哪些时段在用(10 秒一格)和当天的总时长,以及你现在是否在线、用的是哪种客户端(例如 Windows 版、Mac 版、iPhone 版或网站,以及从哪里安装的)。只记录是否在用,不记录你点了什么、输入了什么,也不记录在看哪部视频。我们用这些了解大家怎么使用耳如、改进产品,不用于广告。在线状态只在服务器内存中保留,停止使用几分钟后就清掉;每天的使用时段和时长、用过哪种客户端和最后一次使用的时间保存 90 天。
3.7 访问日志、运行日志和网站统计
- 访问日志:服务器记录访问日志(IP 地址、访问时间、访问的网址、浏览器类型等),用于运维、排查故障和统计访问量。查词时,查的单词和字幕指纹写在网址里,所以也会出现在访问日志中。
- 运行日志:后端程序记录运行情况和错误。日志里的手机号一律打码(例如 138****8000),不记录验证码和密钥;但可能包含账号 ID、字幕指纹、片名、句子开头的片段、加入生词本的单词、查词出错时的单词、学习设置改了哪几项和新的值等,用来排查问题。
- 网站统计:我们用 Umami 统计网站访问情况。Umami 部署在我们自己的服务器上,不是第三方统计服务。按它的默认设置,它记录:访问的网址(含网址里的参数)和页面标题、来源网址、浏览器、操作系统、设备类型、屏幕尺寸、浏览器语言,以及由 IP 地址推算的大致位置(国家、省州、城市)。它不使用 Cookie,也不和你的账号关联。我们也在试用 Matomo 统计网站访问情况,和 Umami 同时运行。Matomo 同样部署在我们自己的服务器上,不使用 Cookie,也不和你的账号关联。它记录的内容和上面相同,另外记录你点开的站外链接和下载、在页面上停留的时间;它保存的 IP 地址去掉了最后两段。
3.8 存在你浏览器里的数据
下面这些存在你这台电脑或手机的这个浏览器里(localStorage、sessionStorage、IndexedDB):
- 登录令牌和账号标记:用来保持登录。浏览器每次请求都会把令牌发给服务器,服务器上也存着一份(见 3.1)。退出登录时清除本机的这份,并通知服务器作废。
下面这些只留在你的浏览器里,不会发给我们:
- 界面和播放偏好(显示方式、布局、播放设置、字幕显示偏好、浮窗位置、登录时选的地区、今日进度的缓存、跟读识别暂时用不了的标记(带账号手机号和日期,只留在这个标签页里,关掉就没了)、是否来过本站);
- 播放记录:每个账号本地视频、NAS 和网盘里的视频各留最近 50 条,记文件名、大小、修改时间、文件类型、最近打开的时间、播放进度、时长、缩略图、选用的字幕轨;本地视频另存浏览器提供的文件句柄(让你下次点一下就能重新打开同一个本地文件),NAS 和网盘的视频只记它是哪个来源里的哪一部,不存播放地址;
- 片库记录:从耳如的片库点开、或在片库里贴网址打开的 B 站 / YouTube 视频,每个账号留最近 50 条,记站点、视频号或网址、片名、封面地址、打开的时间;
- 跟读的原声片段:排对一句时,网页在你的设备上从你自己的视频里剪下这句的几秒声音,存进浏览器,给跟读的「原声」「慢速」用,不上传。换了设备、清了浏览器数据就没有了(跟读改用 AI 读)。
另外,你自己的跟读录音每句留最近读的那一遍在浏览器里,给「听我的」「对比听」回放用,第二天作废。这一份不上传;识别时发出去的那一份见 3.9。
用耳如桌面版时,上面说的「浏览器」就是桌面版本身。另外,你在桌面版里打开 B 站 / YouTube 的视频页时,桌面版会在本机记下这部视频的站点、视频号、片名、封面地址和时间(最多 200 条),用来在观看记录里显示片名和封面。这一份同样不会发给我们。桌面版还在本机存一个随机的安装编号和已经报过哪些字幕对齐结果,安装编号只在上报字幕对齐时用到(见 3.5)。
你可以在浏览器设置里清除本网站的数据来删除它们。我们不使用 Cookie。
3.9 跟读录音和发音
- 跟读录音:每读一遍,网页把这一遍录音发到我们的服务器,服务器交给 AI 服务商 的语音识别,拿回识别出的文字后打分。服务器不保存录音,也不写进日志;只保存识别出的文字、分数和没读准的词(见 3.2)。 在耳如 Mac 版里,登录前要先同意交给 AI 服务商处理,不同意没法登录(见 7)。
- 单词和整句的发音(查词卡、生词本、每日小测里的发音,跟读的「AI 读」):由我们的服务器调用 AI 服务商 的语音合成生成,按内容全站共用缓存——同一个词、同一句只生成一次,存在服务器上,谁听都是同一份。发给 AI 服务商 的只有这个词或这句话的文字,不带你的手机号或账号。某个词的发音还没生成好时,查词卡和生词本会暂时用浏览器自带的朗读功能代替;部分浏览器的朗读会用到浏览器厂商的在线语音服务,这由浏览器厂商按它们自己的政策处理,我们不经手。
3.10 举报记录
在耳如的片库里看哔哩哔哩(B 站)的视频时,你可以举报有问题的视频:内容违规或侵权、画面上有硬字幕、没有字幕或字幕对不上、播放不了,或者其他问题。每次举报,服务器保存:是哪个视频(片库来源和 B 站的视频编号)、举报原因、你写的附言(选填,最多 200 字)、第一次和最近一次举报的时间。登录状态下举报,记在你的账号 ID 下;没登录时,记一个由你的 IP 地址加密钥算出的哈希,不保存 IP 地址本身,这个哈希只用来防止有人刷举报、让同一个人对同一个视频只算一次。举报只用来处理有问题的视频(例如把它从片库下架);必要时,我们会把视频和问题(原因、附言)转告视频所在的平台,但不会附上你的手机号、账号或 IP。
4. 我们不收集什么
- 视频文件本身(画面和声音)不上传,只在你的电脑或手机上播放。
- 不读取你电脑或手机里的其他文件,只读取你主动拖进来或选择的视频。
- 不要求姓名、身份证件号码、住址、通讯录或精确位置。
- 不放广告,不使用第三方广告或跟踪工具。
5. 我们怎么使用这些信息
- 登录和保持登录,确认账号属于持有这个手机号或这个 Google 账号的人;
- 保存和同步你的学习进度、设置和生词本:换一台电脑或手机,用同一个手机号或 Google 账号登录,数据都在;
- 翻译字幕、生成注释、判断句子难度、计分,识别跟读录音并打分,生成单词和整句的发音,执行每日目标;
- 防止滥用:限制验证码的发送频率和次数、限制每天的翻译量、控制费用;
- 处理被举报的视频(例如从片库下架),必要时把问题转告视频所在的平台;
- 记录短信同意,证明你同意过接收验证码短信;
- 保障账号安全(登录记录)、了解各推广渠道带来了多少用户;
- 了解大家每天用耳如多长时间,用来改进产品;
- 排查故障、维护安全、改进功能、了解网站访问量。
我们不出售你的个人信息,不把它用于广告,也不做营销用的用户画像。
6. 短信(SMS)
我们只在你(或你的家人)点了「获取验证码」/「发验证码」时发一条验证码短信,不发营销或推广短信。中国大陆号码由 Spug 推送助手代发(短信里没有「耳如」字样),其他号码由 Twilio 代发(开头写明 Earu AI)。可能产生短信费用;不想再收到,不再请求验证码即可。
7. 我们与谁共享信息
我们只把完成服务所必需的信息交给以下服务商,它们按各自的条款和隐私政策处理这些数据:
- Twilio Inc.(美国):发送短信验证码。Twilio 会收到你的手机号和短信内容(包括验证码本身);用 Twilio 验证服务时,验证码也由 Twilio 生成和核对。
- Spug 推送助手(时巴克公司,中国):给中国大陆手机号发送验证码短信。Spug 会收到你的手机号和验证码(验证码由我们的服务器生成和核对,Spug 只负责发送)。
- Google(美国):登录用途。你选择用 Google 账号登录或绑定时,登录在 Google 的官方页面上完成,Google 按它自己的隐私政策处理你在它那里的账号信息;我们从 Google 收到的是账号的唯一标识、邮箱地址和显示名称(见 3.1)。
- AI 服务商(美国):翻译字幕、生成注释和术语表、判断句子难度、恢复全大写台词的大小写、解析句子、给每日小测标出句子里的词;识别跟读录音和(需要时)字幕校准用的几段视频声音;合成单词和整句的发音。收到字幕句子文字和片名、你每一遍跟读的录音、需要时截下的几段视频声音(见 3.5)、要念的词或句子的文字。我们不向 AI 服务商 发送你的手机号、账号 ID 或其他用来识别你身份的信息;但请注意,片名是从文件名整理出来的(见 3.5),录音里是你自己的声音(见 3.9)。 在耳如 Mac 版里,登录前要先在登录页勾选同意把内容交给 AI 服务商处理(翻译字幕、解析句子、判断句子难度、校准字幕时间、识别跟读录音、合成发音、每日小测)。这些是耳如的核心功能,不同意就没法登录使用;已经登录、但这台 Mac 上没记过同意的,打开耳如时也会先问,不同意就退出登录。同意后这台 Mac 会记住,主窗口和视频网站页面上都算数。想撤回的话,在设置里点「撤回 AI 同意」:撤回后会马上退出登录,不再把内容交给 AI 服务商;想再用,要在登录页重新勾选。
- 腾讯云(Tencent Cloud):提供我们租用的服务器,位于日本东京。服务器端的所有数据——数据库、字幕缓存、日志、网站统计和备份——都存放在这台服务器上。
此外,只有在法律要求时(例如收到有效的法律文书),我们才可能提供相关信息。
8. 保存多久
| 数据 | 保存多久 |
|---|---|
| 账号(手机号或 Google 账号信息、账号 ID、时区)、学习记录(含每日达标、打卡和「最近在播」)、学习设置(含修改历史)、生词本(含已移出的词)、使用量统计 | 账号存在期间一直保存,直到你要求删除 |
| 登录令牌 | 退出登录时通知服务器作废;超过 180 天没用的会被删除;每个账号最多保留 10 个,超出时删掉最久没用的。如果退出时网络不通、服务器没收到通知,那个令牌要等 180 天没用后才删除 |
| 验证码发送记录(手机号、IP、时间) | 2 天后删除 |
| 由我们生成的验证码的哈希 | 10 分钟后失效,2 天后删除 |
| 短信同意记录 | 账号存在期间一直保存,删除账号时一起删除 |
| 登录记录(时间、IP、设备、系统、浏览器) | 每个账号只留最近 50 次;删除账号时一起删除 |
| 推广来源 | 账号存在期间一直保存,删除账号时一起删除 |
| 查词记录 | 90 天后删除;可随时一键清空 |
| 每天的使用时段和时长 | 90 天后删除;删除账号时一起删除 |
| 用过哪种客户端和最后使用时间 | 90 天后删除;删除账号时一起删除 |
| 在线状态(现在是否在线) | 只在服务器内存中,不写入数据库;停止使用几分钟后清掉 |
| 举报记录(视频、原因、附言、时间、账号 ID 或 IP 哈希) | 最近一次举报 1 年后删除;没登录时举报留下的 IP 哈希 90 天后先删除;登录状态下的举报在删除账号时一起删除 |
| 字幕「今天是否打开过」的标记 | 约 2 天后删除 |
| 字幕文字、片名、译文、注释、难度(全站共用缓存) | 长期保存,目前没有自动清理(运营者可以手动删除某一份);删除账号时去掉其中记录的你的手机号 |
| 网盘和 NAS 视频的字幕轨原文(全站共用)及各账号传过、确认过哪一版的记录 | 原文在有账号记录着这一版时保留;每个账号的记录最多 5000 条,随账号一起删除 |
| 片库视频的字幕对齐报告(视频、字幕版本、偏差及所在段落、检测方法、安装编号的哈希) | 最近一次上报 1 年后删除;按账号的每日计数 2 天后删除 |
| 跟读录音 | 不保存:识别完就丢掉(服务器只留识别出的文字、分数和没读准的词,算在学习记录里) |
| 单词和整句的发音(全站共用缓存,不记是谁要的) | 长期保存,目前没有自动清理 |
| 访问日志、运行日志 | 按服务器的日志设置自动轮换删除;目前没有设定固定的保存天数 |
| 网站统计(Umami、Matomo) | 不含账号信息,可能长期保存 |
表里的「X 天后删除」由服务器在运行中顺带清理(例如有人登录、获取验证码或查词时),不是到点立即删除;网站没人使用的时候,过期的数据可能会多留一些时间。
备份:服务器每天凌晨 4:30(东京时间)把数据库、登录和用量文件、字幕缓存备份到同一台服务器上。每份备份保留约 15 天后自动删除,所以已经删除的数据,最多还会在备份里保留约 15 天。
9. 儿童与家长
耳如面向所有学第二语言的人,不是专为儿童设计的。如果孩子使用:
- 账号应当由家长(或其他监护人)持有,并用家长的手机号或 Google 账号登录,用手机号登录时验证码短信发到家长的手机上;孩子在家长同意下使用。
- 跟读要录孩子的声音:每一遍录音只用来识别读得准不准,识别完不保存(见 3.9)。
- 孩子用家长的账号练习时,练习记录、生词本、查词记录会保存在这个账号下。家长可以在网页上看到今日进度、生词本和看视频时查过的词;每一句的练习记录目前网页上看不到,可以发邮件索取。家长也可以随时发邮件要求删除。
- 我们不要求孩子提供姓名、年龄、学校等信息。
- 如果你发现孩子在没有家长同意的情况下用自己的手机号或 Google 账号登录了,请联系我们,我们会删除这个账号的数据。
10. 你的权利
- 查看:学习设置、今日进度、生词本和看视频时查过的词,可以在网页上直接看到。每一句的练习记录、账号信息、使用量等其他数据,目前网页上看不到,可以发邮件索取副本。
- 更正:学习设置可以自己修改。更换手机号等其他更正,请发邮件联系我们。
- 删除:查词记录可以随时一键清空,单词可以随时移出生词本。删除整个账号:手机号账号,在安卓手机或平板上,或在从 Mac App Store 安装的耳如 Mac 版里登录后打开设置,点最下面的「删除账号」,用发到这个手机号的验证码确认;只用 Google 登录的账号,在桌面版里(Windows 版,或从 earu.ai 下载的 Mac 版)登录后打开设置,点「删除账号」,通过重新登录 Google 确认。确认后立即删除,不能撤回(步骤见《删除账号》)。其他设备上没有这个按钮(安卓版和 Mac App Store 版里没有 Google 登录);用其他设备或收不到验证码的,可以发邮件到 support@earu.ai,写明要删除的手机号(只用 Google 登录的,写明登录用的 Google 邮箱),核实你是这个号码或这个账号的使用者后由我们删除。两种方式都会删除账号、登录令牌、学习记录(含每日达标、打卡和「最近在播」)、学习设置、生词本、查词记录、使用量统计、验证码发送记录、短信同意记录、登录记录和登录状态下提交的举报记录,以及保存的 Google 账号标识、邮箱地址和显示名称,并去掉字幕缓存里记录的你的手机号;在设置里自己删除时,还会清掉那台设备上这个账号的最近播放和跟读录音。备份里的副本会随备份轮换删除,最多约 15 天。
- 退出登录:头像菜单里的「退出登录」会清除本机的令牌,并通知服务器作废它(如果当时网络不通,服务器上的令牌会在 180 天没用后删除)。
我们会尽快答复,一般在 30 天内。
11. 安全措施
- 浏览器和网站之间使用 HTTPS 加密连接。
- 验证码有发送频率和次数上限,输错太多次会暂停验证。
- 由我们服务器生成的验证码只保存加密钥的哈希,不保存验证码本身。
- 日志里的手机号打码,不记录验证码和密钥。
- 服务器上的数据目录和密钥文件设置了访问权限,只有运行服务的系统账号和服务器管理员能读取。
- 调用第三方服务用的密钥只保存在服务器上,不会发到浏览器。
没有任何系统能保证绝对安全。登录令牌保存在浏览器里,在共用的电脑或手机上用完请退出登录。如果发生可能影响你的数据的安全事件,我们会在网站上说明情况。
12. 数据存放地点与跨境传输
- 我们的服务器位于日本东京(腾讯云)。无论你在中国大陆、美国还是其他地方使用,服务器端的数据都存放在日本。
- 翻译、解析句子、给每日小测标出句子里的词、识别跟读录音、校准字幕时间(需要时,见 3.5)、生成发音时,字幕文字、片名、跟读录音、截下的几段视频声音和要念的文字会传给美国的 AI 服务商 处理;短信验证码经美国的 Twilio 发送,发给中国大陆号码的经中国的 Spug 推送助手(时巴克公司)发送——这时你的手机号和验证码会传到中国境内。
- 运营者在美国,维护服务器时会从美国远程访问。
使用本服务,即表示你了解你的信息会在上述国家和地区之间传输和处理。如果你不同意,请不要使用本服务。
13. 本政策的变更
如果修改本政策,我们会更新本页顶部的生效日期;有重要变化时,会在网站上提示。
14. 联系我们
- 邮箱:support@earu.ai
- 运营者:耳如运营者(个人开发者,美国)
- 网站:https://earu.ai
另见《服务条款》。
EARU
Privacy Policy
1. Who we are
Earu (Chinese name 耳如, website https://earu.ai) is a video player for listening practice that helps you learn a second language with videos you enjoy. It is available as a website and as desktop apps (including Earu for Mac, installed from the Mac App Store). It is operated by an individual developer (“the operator of Earu”, “we”, “us”) based in the United States. It is not a company.
This policy explains what information we collect when you use the Earu website and service, how we use it, where it is stored, how long we keep it, and how you can access or delete it.
2. Summary
- Your video files play only on your own computer or phone. They are never uploaded to our server.
- To translate, Earu sends the original-language subtitle text of the video, and its title, to our server, which uses our AI provider to translate it into your native language (by default your interface language; you can change it in Settings). Subtitles and translations are shared site-wide by content (see 3.5). In Earu for Mac, you need to agree to this AI processing before you can sign in (see 7).
- In Read-aloud practice, every recording you make is sent through our server to our AI provider to be transcribed into text. Our server does not keep the recordings; it keeps only the transcribed text, the score, and the words you did not say clearly. The copy of your recording used for playback, and the few seconds of original audio cut when you sort a sentence correctly, stay only in the browser on your device (see 3.8 and 3.9). In Earu for Mac, you need to agree to this AI processing before you can sign in (see 7).
- You can sign in with a mobile phone number and an SMS verification code, or with a Google account (desktop app). We send verification codes only, never marketing messages.
- We do not sell or rent your personal information, and we never share it with third parties for marketing.
- Phone-number account: on an Android phone or tablet, and in Earu for Mac installed from the Mac App Store, you can delete your account and all your data yourself at any time in Settings, confirmed with a text code. Account that uses only Google sign-in: in the desktop app (the Windows app, or the Mac app downloaded from earu.ai) you can likewise delete it yourself in Settings, confirmed by signing in to Google again. See Delete your account. You can also email us at support@earu.ai and we will delete it.
3. Information we collect
3.1 Sign-in data (phone number, Google account)
- Phone number: the number you enter to sign in. Mobile numbers from any country or region can be used; numbers that cannot receive text messages, such as toll-free, premium-rate and satellite numbers, are not supported. For an account that signs in with a phone number, the phone number is the account.
- Google account information: only if you sign in with a Google account, or link one in Settings. We receive and store three things from Google: the unique identifier of your Google account, your email address, and your display name. They are used only to recognize your account and let you sign in. We cannot see and do not store your Google password, and we do not ask Google for any other permission, such as reading your email or Drive. Sign-in takes place on Google’s official page opened inside Earu, which is covered by Google’s own privacy policy. An existing phone-number account can link or unlink a Google account in Settings; linking simply adds another way to sign in.
- Account ID: a random identifier our server assigns to each account. Your learning records are stored under it.
- Account dates: when the account was created and when you last signed in.
- Time zone: read from your browser the first time you sign in and used to decide when your “day” starts. You can change it in Settings.
- Sign-in tokens: a random string given to your browser after you sign in, so that you stay signed in. Your browser sends the token with every request so that the server knows it is you. The server stores each token with its account (phone number or Google account), creation time, last-used time, and whether that sign-in was verified by SMS.
- Verification code records: each time a code is requested, we record the phone number, the requesting IP address, the time, and the result. Each wrong code entered is also recorded (without the IP address). These records are used only to prevent abuse and to limit how many codes are sent.
- Verification codes: each code can be used once and is valid for 10 minutes. Codes for mainland China numbers are generated by our server, which stores only a keyed hash of the code, never the code itself, and deletes it after 2 days; codes for other numbers are generated and checked by Twilio’s verification service, and we do not store them. Codes are never written to logs.
- SMS consent records: each time a code is sent to you, we record the phone number, the purpose, the version of the consent notice, when and how many times codes were sent, and a truncated IP network (never the full IP address), as proof that you agreed to receive verification code texts.
- Sign-in records: each time you sign in, we record the time, the IP address, and the device type, operating system and browser as reported by your browser; only the 50 most recent sign-ins per account are kept, for account security and troubleshooting.
- Referral source: if you first signed in through a promotion link with a source tag (for example earu.ai/?src=xhs), that source is recorded on your account so we can see how many users each promotion channel brings.
3.2 Learning records
Each time you score a sentence in Sort practice, the server stores a record: the time and date, the subtitle fingerprint and sentence number, the sentence text, the title, the clip start and end time, the sentence difficulty level, points, the number of hints, time spent, the number of loops, and similar details. We also store your daily target and whether (and when) you reached it; your daily check-ins (when you checked in, how — by finishing Daily Quiz, or by reaching your Sort target on days when only Sort was required — and which items the system skipped that day); and “recently playing”: which subtitles you opened recently and when, one entry per person per subtitle. It is used to validate scoring and to count how many people have opened each subtitle (the operator sees only the count).
In Read-aloud practice, the server stores: the sentences picked for each day (chosen from sentences you sorted correctly, with the sentence text, subtitle fingerprint and sentence number, and title); for each attempt, the transcribed text (up to 200 characters), the score, the words you did not say clearly, whether it passed, and the length, size, and loudness of that recording (used to tell whether any sound was recorded), as well as the recording format, the recording window (the longest that attempt could record), how long transcription took, whether this device had the original audio for that sentence, and a note on any error (for example, a transcription timeout); and records of swapped sentences and system skips. The recordings themselves are not stored (see 3.9). In Daily Quiz, the server stores: the words and questions for each day, each answer (the option chosen, whether it was right, whether time ran out, time taken, how many times the audio was played), and each word's progress and next review date; when a question includes a whole sentence, its source (subtitle fingerprint, sentence number, title) is recorded. The “estimated time” in your study plan is worked out from the times in your recent records (in Sort, from the first sentence to reaching the target; in Read-aloud and Daily Quiz, from starting to finishing).
3.3 Settings
Daily target (a score from 50 to 200), loop timing, subtitle display preferences, difficulty filter, difficulty tier (“Starter / Basic / Advanced / Challenge”), Read-aloud settings (pass mark, after how many failed tries in a row “Swap” is offered, recording length), Daily Quiz settings (answer countdown), and time zone. The daily target is stored as “from this day on, the value is …”, so earlier changes are kept; this lets us judge correctly whether each past day reached its target and how much was due that day. The number of Read-aloud sentences and Daily Quiz questions for each day is not stored separately; it is worked out from that day's daily target.
3.4 Word notebook and lookup history
- Word notebook: words you add yourself (the base form and the form you clicked), the date added, and where the word came from (subtitle fingerprint, sentence number, sentence text, title, time point). Removing a word only marks it as removed; the row is kept so that we can recognize words you added before.
- Lookup history: every word you look up while signed in is recorded (while watching, and also when you click a word on the Word Notebook page), with the time, the page, and where the word came from (subtitle fingerprint, sentence number). Lookups are not recorded when you are not signed in. Lookup history is kept for 90 days. The Word Notebook page shows the words you looked up while watching (one line per word, with a count and the last date); “Clear” there deletes all of your lookup history.
3.5 Subtitle text, title, and timing
When you open a video, the website reads the embedded subtitles on your computer or phone and first sends our server a fingerprint computed from the subtitle content, to ask whether a translation already exists. Only if the server does not have these subtitles yet does the website send the subtitle text merged into sentences, the duration of each sentence, and the title. The video itself is never sent.
The title is cleaned up by the website from the file name (or from a title stored inside the video file): for example the show name, year, season, episode, and episode title. If none of these can be recognized, the title is the file name with the extension, bracketed parts, and quality tags removed. If a file name contains something you would not want others to see, rename the file before opening it.
Videos on video sites: when you watch a Bilibili or YouTube video in the desktop app while signed in, Earu sends us the site and video ID to check whether the video is in the Earu library: for library videos, subtitles come straight from our server, and nothing is uploaded; for other videos, Earu reads the subtitle text that the video site provides and, in the same way, first sends the fingerprint and, only if needed, the subtitle text merged into sentences, the duration of each sentence, and the title shown on the video page. We do not download or store the picture or sound of these videos.
Videos in cloud drives and on a NAS: reading the embedded subtitles of a video from a cloud drive or NAS can be slow, so when you are signed in, after the website has read a subtitle track it sends our server the content of that track as it is (the start and end time and original text of each line, which may include subtitle styling codes) together with a file fingerprint (a hash computed from the file size and a small part at the start of the file; it does not include the file name or path), the track details (track number, codec, language, and track name), and the running time. This is shared site-wide: when someone else opens the same file, their website first spot-checks a few parts against their own file and uses it only if it matches. The server records which account uploaded or confirmed which version (at most 5,000 entries per account); these records are deleted together with the account.
The server stores, by fingerprint, the original subtitle text, the title, sentence durations, the AI-generated translation, notes, glossary, and sentence difficulty levels. Each subtitle is translated only once and shared site-wide: when another user opens subtitles with identical content, they see the existing translation, and may see the title that came from the first uploader's file in the “source” of a notebook word. This shared cache records the phone number of the account that first uploaded it (for usage accounting and cost control); only the operator can see it, and it is not shown to other users. We also count how many times each subtitle is opened and when it was last opened (numbers only; the marker used to count “one person, once per day” is deleted after about 2 days).
Subtitle timing calibration: in some videos the subtitles do not line up with the audio (often drifting further as the video goes on), so the player automatically calibrates the timing of the subtitle track in the language you are learning. It first works this out on your device, by detecting where people are speaking in that audio track and comparing it with the subtitle times. This step uploads no audio; the result (a few numbers) is saved to your account together with the subtitle fingerprint and the video's duration, file size, and audio codec, and is used only for you. Only if the on-device check is not confident does the website cut a few short excerpts from that audio track (usually 5, about 1 minute each) and send them to our server, which passes them to our AI provider's speech recognition to get time-stamped text and work out how far the subtitles need to move. These excerpts are discarded after processing; they are not stored and not written to logs. A result worked out this way is shared site-wide by subtitle fingerprint and the video's duration and size, and contains none of your account information. In Earu for Mac, you need to agree to this AI processing before you can sign in (see 7).
Subtitle sync check for library videos: when you watch a Bilibili or YouTube video from the Earu library in the desktop app, Earu works out on your device where people are speaking in the audio and compares this with the subtitle times, to see whether the subtitles are early or late overall. This step uploads no audio. If you are signed in and it is confident that the subtitles are off by 0.5 seconds or more, it sends us the offset in seconds, how confident it is, roughly which parts of the video the offset was measured in, the name of the detection method, which video and which version of the subtitles it is, and a random install ID, so we can fix the library subtitles; the same subtitle version of the same video is reported only once per device. The report records do not contain your account or IP address (your account is used only to count how many reports are made each day, and that count is deleted after 2 days; the server’s access logs still record requests as usual, see 3.7), and the install ID is stored only as a hash with a secret key. Reports are deleted 1 year after the most recent report.
3.6 Usage statistics
For each account, per day: the number of AI calls and tokens used, sentences translated, sentences served from the cache, new subtitles uploaded, and similar counts, plus the time of last use. We use these for cost control and daily limits.
Usage time and online status: while you are signed in, Earu tells our server from time to time whether you are actively using it (the window is in the foreground and you have used it recently, or a video is playing). For each account, the server records which parts of each day you were using Earu (in 10-second slots) and the total for the day, and whether you are online right now and which kind of client you are using (for example the Windows, Mac, or iPhone app or the website, and where it was installed from). We record only whether you are using Earu, not what you click or type, and not which video you are watching. We use this to understand how Earu is used and to improve it, not for advertising. Online status is kept only in the server’s memory and is cleared a few minutes after you stop using Earu; daily usage times, and which kinds of client you have used and when you last used each, are kept for 90 days.
3.7 Access logs, application logs, and site analytics
- Access logs: the server keeps access logs (IP address, time, URL visited, browser type, and similar request details) for operations, troubleshooting, and traffic statistics. When you look up a word, the word and the subtitle fingerprint are part of the URL, so they also appear in the access logs.
- Application logs: our backend logs its activity and errors. Phone numbers in these logs are always masked (for example 138****8000), and verification codes and secret keys are never logged. The logs may contain account IDs, subtitle fingerprints, titles, the beginning of a sentence, a word added to the notebook, a word whose lookup failed, or which settings were changed and their new values, to help us troubleshoot.
- Site analytics: we use Umami to measure website traffic. Umami runs on our own server; it is not a third-party analytics service. With its default settings it records: the URL visited (including URL parameters) and page title, the referring URL, browser, operating system, device type, screen size, browser language, and an approximate location (country, state or province, city) derived from the IP address. It does not use cookies and is not linked to your account. We are also trying out Matomo to measure website traffic, running alongside Umami. Matomo also runs on our own server, does not use cookies, and is not linked to your account. It records the same information as above, plus clicks on links to other sites, downloads, and time spent on a page; the IP addresses it stores have the last two parts removed.
3.8 Data stored in your browser
The following is stored in this browser on this computer or phone (localStorage, sessionStorage, IndexedDB):
- Your sign-in token and account marker, to keep you signed in. Your browser sends the token to our server with every request, and the server also keeps a copy (see 3.1). Signing out clears the copy in this browser and tells the server to revoke the token.
The following stays only in your browser and is not sent to us:
- interface and playback preferences (view, layout, player settings, subtitle display, floating window position, the country code you chose at sign-in, a cache of today's progress, a marker that Read-aloud recognition is temporarily unavailable (with your account's phone number and the date; kept only in this tab and gone when you close it), whether you have visited before);
- playback history: for each account, up to 50 recent videos from local files and up to 50 from NAS and cloud drives, with the file name, size, modified time, file type, when it was last opened, playback position, duration, a thumbnail, and the chosen subtitle track; for local files, also a file handle provided by the browser (so you can reopen the same local file with one click); for NAS and cloud-drive videos, only which source and which video it was, not the playback address;
- library history: for each account, up to 50 recent Bilibili / YouTube videos you opened from the Earu library, including by pasting a link there, with the site, video ID or URL, title, cover image address, and when it was opened;
- original audio clips for Read-aloud: when you sort a sentence correctly, the website cuts the few seconds of that sentence's audio from your own video, on your device, and stores it in the browser for the “Original” and “Slow” buttons in Read-aloud. These clips are not uploaded. On another device, or after you clear the browser's data, they are gone (Read-aloud then uses AI reading instead).
In addition, your own Read-aloud recording of each sentence (only the latest attempt) is kept in the browser for playback (“Play mine”, “Compare”) and discarded the next day. This copy is not uploaded; for the copy sent for transcription, see 3.9.
In the Earu desktop app, “browser” above means the desktop app itself. In addition, when you open a Bilibili or YouTube video page in the desktop app, the app records that video's site, video ID, title, cover image address, and the time on this device (up to 200 entries), so the title and cover can be shown in your watch history. This copy is also not sent to us. The desktop app also keeps a random install ID and a record of which subtitle sync results it has already reported on this device; the install ID is used only when reporting subtitle sync results (see 3.5).
You can delete this data by clearing this site's data in your browser settings. We do not use cookies.
3.9 Read-aloud recordings and pronunciation audio
- Read-aloud recordings: for each attempt, the website sends that recording to our server, which passes it to our AI provider's speech recognition and scores the transcribed text. Our server does not keep the recording and does not write it to logs; it keeps only the transcribed text, the score, and the words you did not say clearly (see 3.2). In Earu for Mac, you need to agree to this AI processing before you can sign in (see 7).
- Pronunciation of words and sentences (in the word card, the word notebook, Daily Quiz, and “AI reading” in Read-aloud): generated by our server with our AI provider's text-to-speech and cached site-wide by content, so each word or sentence is generated only once, stored on the server, and the same audio is served to everyone. The AI provider receives only the text of that word or sentence, never your phone number or account. If a word's audio is not ready yet, the word card and the notebook temporarily use your browser's built-in speech feature instead; some browsers use their vendor's online voices for this, which is handled by the browser vendor under its own policies, not by us.
3.10 Video reports
When you watch a Bilibili video from the Earu library, you can report a problem with it: content that breaks the rules or infringes someone's rights, subtitles burned into the picture, missing or mismatched subtitles, a video that will not play, or something else. For each report, the server stores: which video it is (the library source and the Bilibili video ID), the reason, any note you add (optional, up to 200 characters), and when you first and most recently reported it. If you are signed in, the report is stored under your account ID; if not, we store a hash computed from your IP address with a secret key, and the IP address itself is not stored; this hash is used only to stop reports from being flooded and to count each person only once per video. Reports are used only to deal with problem videos (for example, removing a video from the library); when needed, we may pass the video and the problem (the reason and note) on to the platform that hosts the video, but never your phone number, account, or IP address.
4. What we do not collect
- The video file itself (picture and sound) is never uploaded; it plays only on your computer or phone.
- We do not read other files on your computer or phone, only the video you drag in or choose.
- We do not ask for your name, government ID number, home address, contacts, or precise location.
- No ads, and no third-party advertising or tracking tools.
5. How we use information
- To sign you in, keep you signed in, and confirm that the account belongs to the holder of the phone number or Google account;
- To store and sync your progress, settings, and word notebook, so they are there when you sign in with the same phone number or Google account on another computer or phone;
- To translate subtitles, generate notes, rate sentence difficulty, score practice, transcribe and score Read-aloud recordings, generate pronunciation audio for words and sentences, and apply your daily target;
- To prevent abuse: limiting how often and how many codes are sent, limiting daily translation volume, and controlling costs;
- To deal with reported videos (for example, removing them from the library) and, when needed, to tell the platform that hosts the video about the problem;
- To keep a record that you agreed to receive verification code texts;
- To keep accounts secure (sign-in records) and to see how many users each promotion channel brings;
- To understand how much time people spend using Earu, so we can improve it;
- To troubleshoot, keep the service secure, improve features, and understand site traffic.
We do not sell your personal information, use it for advertising, or build marketing profiles.
6. Text messages (SMS)
We send one verification code text only when you (or a family member) click “Get code” or “Send code”, and we never send marketing or promotional messages. Texts to mainland China numbers are sent by Spug Push Assistant (they do not contain the name Earu); texts to other numbers are sent by Twilio (they start with Earu AI). Message and data rates may apply; to stop receiving texts, simply stop requesting codes.
7. Who we share information with
We give the service providers below only what they need to provide the service. Each processes the data under its own terms and privacy policy:
- Twilio Inc. (United States): sends SMS verification codes. Twilio receives your phone number and the message text (including the code itself); when Twilio’s verification service is used, Twilio also generates and checks the code.
- Spug Push Assistant (the company 时巴克, China): sends verification code texts to mainland China mobile numbers. Spug receives your phone number and the verification code (the code is generated and checked by our server; Spug only delivers it).
- Google (United States): for sign-in. When you choose to sign in with or link a Google account, sign-in takes place on Google’s official page, and Google handles your account information there under its own privacy policy; what we receive from Google is the unique identifier of the account, the email address, and the display name (see 3.1).
- AI provider (United States): translates subtitles, writes notes and glossaries, rates sentence difficulty, restores normal capitalization for all-caps lines, explains sentences, and picks out the words in each sentence for Daily Quiz; transcribes Read-aloud recordings and, when needed, short audio excerpts used to calibrate subtitle timing; and generates pronunciation audio for words and sentences. Receives the subtitle sentences and the title, each of your Read-aloud recordings, those audio excerpts from the video when needed (see 3.5), and the text of the words or sentences to be spoken. We do not send the AI provider your phone number, account ID, or anything else used to identify you; note, however, that the title is cleaned up from your file name (see 3.5), and that the recordings contain your own voice (see 3.9). In Earu for Mac, before you sign in you need to tick a box on the sign-in page agreeing to have content processed by our AI provider (subtitle translation, sentence explanations, sentence difficulty, subtitle timing calibration, Read-aloud recognition, pronunciation audio, Daily Quiz). These are Earu's core features, so if you don't agree you can't sign in and use the app; if you are already signed in but this Mac has no record of your agreement, we ask when you open Earu, and if you don't agree you are signed out. Once you agree, this Mac remembers it, in the main window and on video sites alike. To withdraw, choose “Withdraw AI consent” in Settings: you are signed out right away and no more content is sent to the AI provider; to use Earu again, tick the box on the sign-in page again.
- Tencent Cloud: provides the server we rent, located in Tokyo, Japan. All server-side data (the database, subtitle cache, logs, site analytics, and backups) is stored on this server.
We may also disclose information when required by law (for example, in response to a valid legal order).
8. How long we keep information
| Data | How long |
|---|---|
| Account (phone number or Google account information, account ID, time zone), learning records (including daily results, check-ins, and “recently playing”), settings (including their change history), word notebook (including removed words), usage statistics | As long as the account exists, until you ask us to delete it |
| Sign-in tokens | Signing out tells the server to revoke the token; tokens unused for more than 180 days are deleted; at most 10 per account (the least recently used is removed). If the network is down when you sign out and the server is not told, that token is deleted after 180 days without use |
| Verification code records (phone number, IP, time) | Deleted after 2 days |
| Hashes of codes generated by our server | Invalid after 10 minutes; deleted after 2 days |
| SMS consent records | As long as the account exists; deleted together with the account |
| Sign-in records (time, IP, device, operating system, browser) | Only the 50 most recent per account; deleted together with the account |
| Referral source | As long as the account exists; deleted together with the account |
| Lookup history | Deleted after 90 days; you can clear it at any time |
| Daily usage times and totals | Deleted after 90 days; deleted together with the account |
| Kinds of client you have used, and when you last used each | Deleted after 90 days; deleted together with the account |
| Online status (whether you are online right now) | Kept only in the server’s memory, never written to the database; cleared a few minutes after you stop using Earu |
| Video reports (video, reason, note, times, account ID or IP hash) | Deleted 1 year after the most recent report; the IP hash from a report made while signed out is deleted first, after 90 days; reports made while signed in are deleted together with the account |
| “Opened today” markers for subtitles | Deleted after about 2 days |
| Subtitle text, titles, translations, notes, difficulty levels (shared cache) | Kept long-term with no automatic cleanup at present (the operator can delete an entry by hand); your phone number is removed from it when your account is deleted |
| Original subtitle tracks of cloud drive and NAS videos (shared site-wide) and records of which version each account uploaded or confirmed | The text is kept while any account’s record points to that version; at most 5,000 records per account, deleted together with the account |
| Subtitle sync reports for library videos (video, subtitle version, offset and where it was measured, detection method, hash of the install ID) | Deleted 1 year after the most recent report; the per-account daily count is deleted after 2 days |
| Read-aloud recordings | Not stored: discarded once transcribed (the server keeps only the transcribed text, score, and unclear words, as part of your learning records) |
| Pronunciation audio for words and sentences (shared cache; not linked to who requested it) | Kept long-term with no automatic cleanup at present |
| Access logs and application logs | Rotated out automatically under the server's log settings; no fixed number of days is set at present |
| Site analytics (Umami, Matomo) | Contains no account information; may be kept long-term |
“Deleted after N days” in this table means the server removes expired data during routine cleanup while it runs (for example when someone signs in, requests a code, or looks up a word), not at the exact moment it expires. When nobody is using the site, expired data may stay somewhat longer.
Backups: every day at 4:30 a.m. (Tokyo time) the server backs up the database, the sign-in and usage files, and the subtitle cache to the same server. Each backup is deleted automatically after about 15 days, so deleted data may remain in backups for up to about 15 days.
9. Children and parents
Earu is for anyone learning a second language and is not designed specifically for children. If a child uses it:
- The account should be held by a parent (or other guardian), who signs in with the parent's phone number or Google account, so that, with a phone number, verification codes go to the parent's phone; the child uses the service with the parent's permission.
- Read-aloud practice records the child's voice: each recording is used only to check how well the sentence was read and is not kept after transcription (see 3.9).
- When a child practices under a parent's account, practice records, the word notebook, and lookup history are stored under that account. The parent can see today's progress, the word notebook, and the words looked up while watching on the website; per-sentence practice records are not shown on the website at present, but the parent can email us for them. The parent can also ask us by email to delete them at any time.
- We do not ask children for their name, age, school, or similar information.
- If you learn that a child signed in with their own phone number or Google account without a parent's permission, contact us and we will delete that account's data.
10. Your choices and rights
- Access: settings, today's progress, the word notebook, and the words you looked up while watching are visible on the website. Per-sentence practice records, account details, usage, and other data are not shown on the website at present; email us for a copy.
- Correction: you can change your settings yourself. For other corrections, such as changing the phone number, email us.
- Deletion: you can clear your lookup history and remove words from the notebook at any time. To delete your whole account: for a phone-number account, on an Android phone or tablet, or in Earu for Mac installed from the Mac App Store, sign in, open Settings, tap “Delete account” at the bottom, and confirm with a code texted to the account’s phone number; for an account that uses only Google sign-in, in the desktop app (the Windows app, or the Mac app downloaded from earu.ai), sign in, open Settings, tap “Delete account”, and confirm by signing in to Google again. Once you confirm, the account is deleted right away and this cannot be undone (see Delete your account). This button is not available on other devices (the Android app and the Mac App Store version have no Google sign-in); if you use another device or cannot receive the code, email support@earu.ai with the phone number (or, for an account that uses only Google sign-in, the Google email address you signed in with); once we confirm that you are the user of that number or account, we will delete it for you. Either way, we will delete the account, sign-in tokens, learning records (including daily results, check-ins, and “recently playing”), settings, word notebook, lookup history, usage statistics, verification code records, SMS consent records, sign-in records, and reports you made while signed in, as well as the Google account identifier, email address and display name we stored, and remove your phone number from the shared subtitle cache; deleting in Settings also clears this account’s recently played videos and read-aloud recordings on that device. Copies in backups are deleted as the backups rotate, within about 15 days.
- Signing out: “Sign out” in the avatar menu clears the token in this browser and tells the server to revoke it (if the network is down at that moment, the server deletes the token after 180 days without use).
We will respond as soon as we can, normally within 30 days.
11. Security
- Connections between your browser and the website use HTTPS.
- Verification codes are rate-limited, and verification is paused after too many wrong codes.
- For codes generated by our server, only a keyed hash is stored, never the code itself.
- Phone numbers are masked in logs; verification codes and secret keys are never logged.
- On the server, the data directory and the key file have restricted permissions: only the system account that runs the service and the server administrator can read them.
- Keys for third-party services are kept on the server and never sent to browsers.
No system is perfectly secure. Your sign-in token is stored in your browser, so sign out when you finish on a shared computer or phone. If a security incident may affect your data, we will post a notice on the website.
12. Where data is stored and international transfers
- Our server is in Tokyo, Japan (Tencent Cloud). Wherever you use the service (mainland China, the United States, or elsewhere), server-side data is stored in Japan.
- For translation, sentence explanations, picking out words for Daily Quiz, Read-aloud transcription, subtitle timing calibration (when needed, see 3.5), and pronunciation audio, subtitle text, titles, Read-aloud recordings, audio excerpts from the video, and the text to be spoken are sent to our AI provider in the United States; SMS codes are sent through Twilio in the United States, and codes for mainland China numbers through Spug Push Assistant (the company 时巴克) in China, in which case your phone number and the code are transferred to mainland China.
- The operator is in the United States and accesses the server remotely from there for maintenance.
By using the service, you understand that your information is transferred to and processed in these countries. If you do not agree, please do not use the service.
13. Changes to this policy
If we change this policy, we will update the effective date at the top of this page, and we will show a notice on the website for important changes.
14. Contact us
- Email: support@earu.ai
- Operator: the operator of Earu (individual developer, United States)
- Website: https://earu.ai
See also our Terms of Service.