# On-device live captions: sound and text stay on this computer / 本机同传字幕：语音和文字不离开这台电脑

> Source: https://listen.365ui.com/live-captions
>
> This computer listens to the room, recognises speech on its own CPU and translates in the browser, and shows bilingual captions on the screen. The sound, the original and the translation stay on this computer, reaching neither us nor any cloud service, for meetings whose content must stay in the room. Free once you sign in, with no limit; the paid version takes away the badge and the pauses.

## On-device live captions: sound and text stay on this computer

This computer listens to the room, recognises speech on its own CPU and translates in the browser, and shows bilingual captions on the screen. The sound, the original and the translation stay on this computer, reaching neither us nor any cloud service, for meetings whose content must stay in the room. Free once you sign in, with no limit; the paid version takes away the badge and the pauses.

### Two languages

One row is what the speaker says, the other its translation. The second can be any language; when it is Chinese, English, Japanese or Korean too, the speaker can switch between the two and each sentence goes in its own row.

The first start downloads a 240 MB recognition model; this browser keeps it, so later starts do not.

Sign in with email or Google; nothing to pay. It only tells us who is using it: sound and text still never leave this computer.

The free version asks every 45 minutes: still there?

### The paid version: no badge, no pauses

$5 a month, or $50 a year, paid once; it does not renew by itself. What it buys is a screen with no “Free version” badge that never pauses to ask; recognition and translation still run on this computer and are exactly as accurate as the free version.

### Accuracy, and how it differs from paid interpretation

- The free and paid versions use the same recognition and translation and are exactly as accurate as each other, and both less so than the $25-per-room-hour captions service: of the 13 English names and terms in the same 13-minute Chinese talk, on-device live captions get 5 right, and the captions service 11 with no glossary, and it can follow yours.
- On-device live captions: one screen, on-device recognition of Chinese, English, Japanese, Korean and Cantonese, the browser’s own translation, no glossary, accuracy not guaranteed; sound and text stay on this computer.
- Paid interpretation: attendees read captions and hear interpretation on their own phones, 14 languages, your names and terms translated your way, transcripts in every language afterwards, a standby engine, and a pro-rata refund if an outage runs past the agreed limit; the sound is processed in the cloud.

- [See paid interpretation pricing](https://listen.365ui.com/pricing/en)
- [Try paid interpretation free](https://listen.365ui.com/trial)

Recognition: SenseVoiceSmall by FunASR/FunAudioLLM (FunASR model licence); engine: sherpa-onnx (Apache-2.0); translation: the browser’s on-device translator. Provided as is. [Model licence](https://github.com/modelscope/FunASR/blob/main/MODEL_LICENSE) · [Engine licence](https://listen.365ui.com/live-captions/engine/LICENSE-sherpa-onnx.txt)

## 本机同传字幕：语音和文字不离开这台电脑

这台电脑听会场的声音，在本机识别、在浏览器里翻译，双语字幕直接投到屏幕上。语音、原文和译文都只在这台电脑上，不发给我们，也不发给任何云端服务，适合不能把内容传出去的会。登录后免费用，不限次数；付费版去掉角标和暂停。

### 两种语言

一行是讲者说的话，另一行是译文。第二种可以选任何一种语言；它也是中、英、日、韩时，讲者两种来回换都行，每句话进它自己那一行。

第一次开始要下载约 240 MB 的识别模型，之后存在这个浏览器里，再开不用下载。

用邮箱或 Google 登录，不用付钱。登录只是让我们知道谁在用，声音和文字照样不离开这台电脑。

免费版每 45 分钟问一次：还在用吗？

### 付费版：去掉角标和暂停

每月 $5，或每年 $50，一次付清，不自动续费。买的是大屏上没有「免费版」角标、也不会每隔一段时间暂停；识别和翻译还在这台电脑上，准确率和免费版完全一样。

### 准确率，以及和付费同传的差别

- 免费版和付费版用的是同一套识别和翻译，准确率完全一样，但都不如每房间小时 $25 的字幕服务：同一段 13 分钟中文演讲里夹着的 13 个英文人名和术语，本机同传字幕认对 5 个，后者不给术语表也认对 11 个，还能按您的术语表翻。
- 本机同传字幕：一块屏，本机识别中、英、日、韩、粤，浏览器自带的翻译，没有术语表，准确率不保证；语音和文字不离开这台电脑。
- 付费同传：观众用自己的手机看字幕、听译音，14 种语言，按你的人名和术语表翻译，会后各语言文字稿，有备用引擎，中断超过约定按比例退款；声音要传到云端处理。

- [看付费同传价格](https://listen.365ui.com/pricing/zh)
- [免费试用付费同传](https://listen.365ui.com/trial)

识别模型 SenseVoiceSmall by FunASR/FunAudioLLM（FunASR 模型开源协议），运行库 sherpa-onnx（Apache-2.0），翻译用浏览器自带的本机翻译。按现状提供。 [模型协议](https://github.com/modelscope/FunASR/blob/main/MODEL_LICENSE) · [运行库许可](https://listen.365ui.com/live-captions/engine/LICENSE-sherpa-onnx.txt)
