I stopped reading subtitles and started listening — notes on real-time AI dubbing in a browser tab

For a long time I treated foreign-language video like a reading assignment.

Open the tab. Turn on captions if they exist. Pause when the line is too fast. Scrub back. Open a dictionary. Miss the next joke anyway.

That works for a recorded lecture you control. It falls apart for a livestream, a Coursera week you are already behind on, or a meeting where people keep talking while you are still parsing the last sentence.

Subtitles help you decode. They do not always help you stay present. At some point I wanted the opposite posture: keep my eyes on the video, and let my ears take the translation.

So I started building toward real-time AI dubbing on whatever audio is already playing in the browser — not “upload a file and wait for an MP4.”

The job is not “translation quality” in the abstract

Offline tools optimize for a clean transcript and a polished voiceover. Live tools optimize for a different constraint set:

1. Stay on the same tab — no export, no special player
2. Latency low enough to feel watchable — for me, roughly half a second is the line between tool and distraction
3. Original + translation available together when I want them (learning mode)
4. A spoken translation when I am tired of reading
5. Fail gracefully when the network blips — reconnect without throwing away the session mental model

If you only chase BLEU scores, you can still ship something people rage-quit in minute three of a Twitch stream.

Tab audio is a blunt but useful boundary

Site-by-site integrations never end. YouTube today, Coursera tomorrow, some random course player next week.

For v1, “if the tab is playing sound” is a blunt MVP line that actually ships. Capture the tab’s audio, chunk it sensibly (silence is not worth uploading), show captions, and — when dubbing is on — play a translated voice while optionally ducking the original.

That last part matters more than I expected. A dubbed voice at 100% with the original also at 100% is mush. Ducking the source while the AI voice speaks is a small control with a large “this feels intentional” effect.

Dual subtitles changed how I learn; dubbing changed how I watch

Bilingual captions (source + target) cut how often I alt-tab to a dictionary. I still use them when I want accuracy or vocab.

Dubbing is for a different night: long stream, tired eyes, I mostly want to hear the meaning and glance at the picture. Same pipeline, different output surface.

I also learned that “subtitle style” is not cosmetic. A cinema burn-in and a side panel transcript are different reading postures. One keeps you in the video; the other is closer to notes.

What I would tell myself before building

1. Solve the pause loop, not “AI dubbing” as a vibe.

The job-to-be-done was staying present on live or long-form speech.

2. Universal tab audio beats perfect site hacks for v1.

Special-case forever and you never ship Twitch.

3. Voice choice and mix are product, not polish.

People abandon tools that sound wrong even when the text is fine.

4. Desktop is a different problem.

Browser tab capture covers most courses and streams. Downloaded files, desktop Zoom, and “everything on the PC” need a different listener. I ended up with a Windows 11 app for that case — same account, same hours — because the extension alone cannot hear what never enters a tab.

The thing I actually use

I packaged this workflow as DubTab — Chrome/Edge extension, optional Windows app, freemium minutes.

Site: dubtab.com

Chrome Web Store: DubTab

Short demo: YouTube

I am not doing a pricing pitch here. The interesting part for me was treating live dubbing as a latency + mix + session problem, not only an ASR problem.

Open questions

• Where do you split live captions vs archival transcripts in product and storage?
• How do you explain tab-audio permissions without sounding like malware?
• For learners: default to bilingual captions, dubbing, or both — what feels least noisy on a 13" laptop?

If you have built realtime speech tools — or you have your own pause-rewind ritual for foreign streams — I would love to hear what you landed on in the comments.

---

Built this because I needed it. Feedback: [e-mail]