Every video speaks your language.

Click play — subtitles are already there. Not machine translation: every line is re-cut to broadcast subtitle rules.

Desktop Chrome and Edge (install from the Chrome Web Store). Works on YouTube.

A video player with a Chinese subtitle line above the original English line, as Earfish shows it
Illustration of how Earfish shows subtitles. The subtitle lines are real Earfish output.

Two reasons

Fast, and it reads like subtitles

Fast: no waiting

Earfish reads the captions the video already has and translates the first lines before anything else. On a video that nobody has opened before, the first translated line appeared after

5.8s (median)

Measured 2026-10-19: 6 TED talks never opened before, local test server, time from opening the video to the first translated line. Videos other people have watched come from the cache.

Good: not machine translation

Each line is re-cut to subtitle rules: how long a line is, whether you can read it in time, how long it stays on screen. A line stays at least 1.3 s. Lines that only flash by are now

0.7% of lines

A/B test 2026-09-27 on 10 held-out videos: lines shorter than 1.3 s fell from 7.2 % to 0.7 % after the rule shipped.

Same lines, side by side

Machine translation vs Earfish

The auto captions of this video mishear two product names. Line-by-line machine translation keeps the errors. Earfish fixes the English first, then translates.

Auto captions (original)Line-by-line machine translationEarfish
We're going to compare between Codex, Cloud Code, Cursor, and Devon. 我们将比较 Codex、Cloud Code、Cursor 和 Devon。 我们将比较Codex、Claude Code、Cursor和Devin
Codex having the GPT models, Cloud Code having the Claude models. Codex 具有 GPT 模型,Cloud Code 具有 Claude 模型。 Codex用GPT模型 Claude Code用Claude模型
That's why in today's video, we're going to talk about Codeex. 这就是为什么在今天的视频中,我们将讨论 Codeex。 所以今天的视频 我们来聊聊Codex

Source: YouTube video 63p9hdRRXkM (a review of AI coding tools). Left: the auto captions Earfish received on 2026-08-09. Middle: the same lines sent one by one to a general online machine translation service (2026-10-20). Right: Earfish output from a head-to-head test the same day (2026-08 pipeline). Red = misheard name, green = fixed by Earfish. Earfish drops end punctuation, following subtitle rules.

Questions

Is it free?

Free to install. Each install comes with a free allowance of video time. No sign-up needed.

What do you do with my data?

To translate, Earfish sends the video page address, title and captions to our server, with a random install ID. It does not read your passwords or cookies, and it does not look at pages other than the video sites it supports. When you turn dubbing on, it also sends the audio of the video. Details are in the privacy policy. Privacy policy

Which sites and languages work?

YouTube, in desktop Chrome and Edge. The video needs captions (most do). You can translate into 8 languages. The default is your browser’s language; change it in the fish menu of the player.

Put a fish in your ear.