支持

多国语音互译 · VoiceTranslator

这是一个完全离线的语音翻译应用:录音不出设备,译文不用等网络。 下面是最常见的问题。若没有解决你的疑问,请直接发邮件联系。

它真的不需要联网吗?

是的。应用内没有任何网络请求——没有账号、没有登录、没有统计上报。 你可以打开飞行模式验证:录音、识别、翻译、朗读全部照常工作。

哪些语言可以直接用?

装好即可使用:中文 ↔ 英文、中文 ↔ 日文。 对应的语音识别模型与翻译模型已经内置在安装包里。

提示「缺少模型包」怎么办?

英文 ↔ 日文、韩语、粤语等语向需要一份更大的通用翻译模型。为了把安装包控制在 合理体积,它没有内置,需要你按需导入一次:

  1. 点击右上角齿轮进入设置;
  2. 找到模型包一栏,它会明确指出缺的是哪一个;
  3. 点导入模型包,从「文件」App 或隔空投送中选取 .vtmodel 文件。

导入后模型只保存在本机沙盒内,不会上传;不需要时可在设置页移除,以回收存储空间。

录音会被保存或上传吗?

都不会。音频只存在于内存中,用于当次识别,处理完立即丢弃,既不写磁盘也不离开设备。

为什么第一次翻译要等一会儿?

切换语向时模型需要重新载入并做一次预热(大约 10 秒左右),之后连续翻译就是实时的。 这是离线推理的正常代价——换来的是数据不出设备。

朗读没有声音?

请先检查设备是否处于静音模式,以及系统「设置 → 辅助功能 → 朗读内容」中的语音是否可用。 应用在朗读期间会暂停录音(系统只允许一路音频),朗读结束后自动恢复。

支持哪些设备?

iPhone 与 iPad,系统版本 iOS 15.0 及以上。

还需要帮助?

发邮件到 zhangxuefeng@batonsoft.com, 描述你的设备型号、系统版本和具体现象,我们会尽快回复。


Support

VoiceTranslator — Offline Speech Translation

This app is fully offline: your audio never leaves the device and translations never wait for a network.

Does it really need no network?

Correct. There are no network requests of any kind — no account, no login, no analytics. Turn on Airplane Mode to verify: recording, recognition, translation and read-aloud all keep working.

Which languages work out of the box?

Chinese ↔ English and Chinese ↔ Japanese work immediately after install; their speech and translation models ship inside the app.

It says a model package is missing

English ↔ Japanese, Korean and Cantonese need a larger general-purpose translation model. It is not bundled, to keep the download small. To add it:

  1. Tap the gear icon to open Settings.
  2. Find the Model packages section — it names the missing one.
  3. Tap Import model package and pick a .vtmodel file from Files or AirDrop.

Packages stay inside the app sandbox, are never uploaded, and can be removed at any time to reclaim storage.

Are recordings stored or uploaded?

Neither. Audio exists in memory only for the current utterance and is discarded immediately — never written to disk, never sent anywhere.

Why is the first translation slow?

Switching language direction reloads and warms up the model (about ten seconds); after that, continuous translation is real-time. That is the normal cost of on-device inference.

Which devices are supported?

iPhone and iPad running iOS 15.0 or later.

Contact

zhangxuefeng@batonsoft.com — please include your device model, OS version and a description of the problem.