Overhear turns any audio playing on your device into live subtitles in the language you choose: a video in a browser tab, a livestream, a media player, or a file you open. Other tools only translate videos that already carry captions, or stay locked inside one meeting app. Overhear needs neither. It listens to the raw sound, detects the language on its own, and reads the meaning in context, so a likely mishearing becomes what the speaker actually meant instead of nonsense. You see the translation live with the original speech beneath it. Open a file and it plays right inside Overhear, so you can scrub or rewind while every line stays stamped to the exact moment it was spoken in the clip.

Free trialYou have 1:30 of live translation. The clock starts when you pick a source below.

Auto-detect follows the speaker between languages mid-sentence. Pin a language only when the audio is entirely in that language.

Examples: a crypto or markets livestream, a football or soccer match, a news broadcast, a geology or nature documentary, a cooking show, a tech keynote, a movie or TV episode. Naming the subject sharpens both the hearing and the translation. Leave it blank to auto-detect.

Audio options

On by default. A few seconds after a sentence the recogniser was unsure about, a non-streaming engine re-hears it and the transcript line (and its translation) update if it did better. Sentences it was confident about are left as they are. Live subtitles are never delayed.

On by default. A file you open is transcribed whole before it plays, so the full translation is ready up front, every line is stamped to the exact clip position, and it costs less than listening live. Turn off to stream a file live instead.

On by default. A speech detector decides when nobody is talking and pauses the stream, so you are not billed for silence or music between sentences. It waits a moment before pausing and resumes on the first word, so speech is not cut. Applies from the next session.

On by default. Lifts quiet captured audio (often much quieter than what you hear) to a good level for translation, without pumping up background noise. Turn it off if a clip sounds over-processed.

Off by default. Runs a neural noise suppressor (RNNoise) on the audio before recognition. Try it on noisy clips (crowds, traffic, music beds). Takes effect at the next session start. Not yet measured on the test corpus, so compare with it off.

Off by default. Narrows to the speech range to favour voices over music and hiss. Try it only if a clip is hard to hear: it can occasionally hurt accuracy.

Advanced: sentence timing

How long a pause has to be before a sentence closes and gets translated. Shorter values feel faster but chop sentences; longer values wait for whole thoughts.

Recogniser values apply at the next session start; the settle pause applies now.
Remembered fixes0 fixes

Each fix teaches the next session: misheard words are sent to the recogniser as priority terms, exact phrases are corrected as they arrive, and preferred translations are passed to the translator.

No fixes yet. Click Fix on any line of the transcript to correct what was heard or how it was translated. Every fix is remembered and applied to future sessions.

Choose a source

In the picker that opens, choose the tab that is playing and tick “Also share tab audio”. Without the tick we receive silence.