Free · private · in your browser
Add music, review the detected speech, then export the balanced mix.
Ducking is the reason background music in a well-made video never competes with the narration: the music drops automatically the moment someone speaks and comes back up when they stop. This tool detects speech in your soundtrack, lowers the music underneath it by an amount you choose, and lets you review the detection before exporting.
Doing this by hand means drawing a level change at every sentence, and a five-minute narration can contain a hundred of them. Automatic ducking derives the moves from the speech itself, so every phrase gets the same treatment, and adding or trimming narration does not mean redrawing everything. The result is also more consistent than hand-drawn automation, which tends to vary as concentration fades over a long edit.
Between 6 and 12 dB is the usual range. Less than that and the music still competes with quiet consonants; much more and the track disappears entirely, so the ducking becomes the thing you notice instead of the mix. Timing matters just as much: the music should drop quickly enough that the first word is not buried, and recover slowly enough that it does not lurch up in the gap between two sentences. A slower recovery, around half a second, sounds natural; an instant one sounds mechanical.
Speech detection can be fooled in both directions. A loud passage in the music can read as speech and duck the track against itself, while quiet or distant narration can be missed, leaving music over the words you most needed to hear. Reviewing the detected regions before export catches both, and it is much faster than listening to the whole mix hunting for the moment it went wrong. Ducking also cannot rescue a mix where the music was simply too loud to begin with — set a sensible starting balance first.
6–12 dB covers most cases. Less leaves music competing with quiet consonants; more makes the ducking itself the thing listeners notice.
A loud passage in the music was detected as speech. Review the detected regions before exporting and adjust the sensitivity so the track stops triggering itself.
Around half a second. Faster sounds mechanical and lurches between sentences; much slower leaves the music suppressed through the gaps where it should be carrying the video.
Ducking needs two elements — a voice and a bed to lower. If the music is already mixed into a single track, it cannot be separated out and ducked independently.
Yes, for anything with narration. A fixed level is either too loud under speech or too quiet in between; ducking gives you both.
