OBS & streaming tech
How to duck music under translated speech in OBS
Use OBS sidechain compression to lower music during translated playback, with verified starting values, test order, and misrouting checks.
Lab Editorial3 min read
Background music can support a live-selling room and still make translated playback unintelligible. Riding a fader by hand asks an operator to react to every sentence. OBS can do the mechanical part with sidechain compression, commonly called ducking.
Ducking does not improve recognition or translation. It solves a later problem: once translated speech reaches the mix, which signal gets the listener’s attention?
The signals must be separate
OBS needs one source for translated playback and another for music. If ObsTrans already sends speech through a virtual cable into a dedicated Audio Input Capture, the required separation exists. Otherwise, start with the virtual audio cable guide.
Run three checks before adding a filter:
- translated playback moves only the translation meter;
- pausing music makes only the music meter stop;
- disabling playback makes the translation meter fall to zero.
If one source contains both signals, compression settings cannot repair the routing.
Put the filter on the signal that moves down
Open Filters on the music source and add Compressor. In Sidechain/Ducking Source, choose the translated-audio source.
The direction is easy to reverse. The sidechain is the trigger; the source holding the filter is the signal being reduced. Putting Compressor on translation causes the opposite of what you want.
OBS’s official Compressor reference offers a ducking starting point: 32:1 ratio, -36 dB threshold, 100 ms attack, 600 ms release and 0 dB output gain. The page explicitly presents these as values to start from and adjust.
Tune in a fixed order
Adjust Threshold first. If translation is audible but music never falls, the threshold may sit above the trigger’s real level. If music moves while no translation is playing, the translation path may contain noise or the threshold may be too low.
Next adjust Attack. Too slow masks the beginning of each translated sentence; too fast sounds like the music falls through a trapdoor.
Adjust Release last. Extend it when music pumps up between short phrases. Shorten it when music stays reduced long after playback finishes. Leave Output Gain at zero while diagnosing; adding makeup gain works against the purpose of ducking.
Keep the total mix below clipping
Ducking lowers one source but does not guarantee a safe master level. OBS’s Audio Mixer Technical Details explains that final audio must remain below 0 dBFS or the viewer hears clipping. It identifies -20 dBFS as an alignment level near average speech and -9 dBFS as the permitted maximum level.
Treat those as references, not a universal loudness target. Record the actual combination of host, translated speech and music. If you need a ceiling, OBS’s Limiter documentation says the Limiter belongs last in the filter chain.
Where ObsTrans fits
ObsTrans can send translated playback to a virtual cable while keeping a separate local monitor route. That gives OBS a clean source to use as the sidechain trigger, and the built-in output tests help establish that both routes work.
It cannot decide the artistic balance. A product demonstration may need environmental sound to remain prominent; a spoken offer may need music almost absent. That decision belongs to your recording and listener test.
Production acceptance
Record continuous explanation, several short phrases and a long silence. Listen for:
- the first word of every translated sentence;
- pumping during short gaps;
- natural recovery after speech ends.
Only then inspect the platform preview. YouTube’s official streaming tips call for checking Live Control Room before starting and monitoring audio and video continuously. The platform preview is the final observation point, not a substitute for a controlled local recording.
Steps
- 1
Isolate translated playback first
Route the translation app through a virtual cable and confirm that only the translated-audio meter moves during playback. Ducking cannot react selectively when translation is mixed into Desktop Audio.
- 2
Add Compressor to the music source
Open Filters on the background-music source and add Compressor. The filter belongs on the signal that should become quieter, not on translated speech.
- 3
Choose translation as the sidechain
Set Sidechain/Ducking Source to the isolated translated-audio source. If it is absent, verify that the source exists in the scene and is not merely part of Desktop Audio.
- 4
Begin with the documented baseline
Start with OBS guidance of 32:1, -36 dB, 100 ms attack, 600 ms release and 0 dB output gain, then change only one control at a time.
- 5
Accept it from a recording
Record continuous speech, short phrases and a long pause. Approve the mix only when words remain clear, music does not pump between phrases, and recovery sounds natural.
Frequently asked questions
- Which source gets the Compressor filter?
- The source you want to reduce, normally music. Translated speech is selected inside that filter as the Sidechain/Ducking Source.
- Why is translated audio missing from the sidechain list?
- It is probably not an independent OBS source. Route it through a virtual cable into its own Audio Input Capture instead of capturing it as part of Desktop Audio.
- Are the official OBS values production settings?
- No. OBS calls them a starting point. Your threshold depends on actual translated-speech level, while attack and release must be judged against your own speaking rhythm.
KeywordsOBS sidechain compressionaudio duckingtranslated audiolive stream mixingOBS Compressor
