Podcast Music Ducking That Keeps Speech Clear and Even
The most common note I give on a new show is not about the microphone. It is about the music under the host's first sentence.

The most common note I give on a new show is not about the microphone. It is about the music under the host's first sentence. Either the bed sits so loud that listeners on a bus lose every third word, or it drops out so abruptly that it sounds like someone tripped over a cable. Both problems come from the same place: ducking that was set once, by ear, on studio headphones, and never checked anywhere else.
Ducking is just lowering music while someone talks and bringing it back when they stop. Doing it well is mostly about three numbers, how far down, how fast, and how fast back up, plus a willingness to draw a few lines by hand. This is the method I use on interview shows, with the settings I start from in Reaper, Audition and Descript.
Why music fights speech harder on phones
Most people hear podcasts on earbuds, phone speakers or a car stereo with road noise underneath. All three lose low end and compress what is left. A music bed that felt like a gentle cushion on closed-back headphones turns into a wall in the 1 to 4 kHz range, which is exactly where consonants live. The listener does not think "the music is too loud." They think the host mumbles.
That is why I judge a bed by how much it masks speech, not by how loud it sounds in isolation. A busy acoustic guitar part at minus 25 dB can be more distracting than a soft synth pad at minus 18 dB, because the guitar's attack sits right on top of the voice. If you pick your own music, choose beds with little going on between 1 and 4 kHz, or plan to cut that range with EQ on the music track only.
The settings I start from
Every show needs its own adjustment, but starting points save an hour. These assume your finished episode targets minus 16 LUFS integrated for stereo, the figure I explain in the guide to podcast loudness targets.
| Setting | Starting value | What it controls | Move it when |
|---|---|---|---|
| Duck depth | 12 to 15 dB below the music's full level | How far the bed drops under speech | Deeper for dense music or soft voices |
| Bed level under speech | About 20 to 24 dB below the voice | Whether words stay clear on earbuds | Lower if consonants smear on a phone speaker |
| Fade down (attack) | 150 to 300 ms | How quickly music gets out of the way | Faster if the first syllable gets covered |
| Fade up (release) | 600 ms to 1.5 s | How the music returns after a phrase | Slower if it pumps between sentences |
| Hold | 400 to 800 ms | How long it stays down after speech stops | Longer for hosts with long pauses |
The release matters more than people expect. A fast return makes the music swell into every breath, and the whole intro starts to breathe like a cheap radio ad. A release of about a second, plus a hold of half a second, keeps the bed down through normal pauses and lets it rise only when the host has actually finished.
Reaper, Audition and Descript in practice
In Reaper I put ReaComp on the music track, set its detector input to auxiliary, and send the voice bus to channels 3 and 4. Ratio around 6:1, threshold low enough to give 12 dB of reduction on normal speech, attack 200 ms, release 1000 ms. In Audition, the Auto Duck feature in the Essential Sound panel does a decent first pass; set Sensitivity near 6, Reduce By 15 dB and Fades at 800 ms, then click Generate Keyframes so you get editable points instead of a black box. Descript has no true sidechain, so I set the music clip's volume by hand at each speech entry, which sounds odd but takes ten minutes for a two minute intro.
Against setting it and forgetting it
A lot of tutorials say a sidechain compressor is the professional way to duck and that keyframes are for beginners. I disagree, at least for spoken word.
A compressor reacts to signal. It does not know the difference between a host starting a sentence and a host clearing their throat, laughing, or saying "um." Each of those dips the music, and over a three minute cold open you get a bed that twitches. A compressor also reacts after the voice arrives, so even with a short attack the first syllable lands on full music.
What works better, in my experience, is using automatic ducking to rough out the shape and then converting it to volume automation you can edit. I move each downward fade so it starts about 100 to 200 ms before the first word, delete the dips caused by breaths and laughter, and flatten the line through short pauses. On a typical intro that is maybe 15 or 20 points. It is the single biggest jump in polish I can offer a show for the time it takes.
Sidechain compression still earns its place on long beds under an entire segment, such as a news roundup with music running for twelve minutes. There, editing every point is unreasonable. Even then I set the release slow, around 1.5 seconds, and accept a little less clarity in exchange for a bed that does not pump.
Checking the mix outside the studio
Ducking decisions made on good headphones are guesses until you hear them on bad playback. My routine before signing off an intro takes about fifteen minutes:
- Export the first five minutes at your final loudness and format, usually 128 kbps MP3 or 96 kbps AAC.
- Listen on the phone speaker held at arm's length. If any word is hard to follow, pull the bed down 2 or 3 dB under that phrase.
- Listen on cheap earbuds while walking or with a kitchen tap running. Road and water noise mimic what commuters actually deal with.
- Listen in a car if you can. Car stereos boost bass and mids, which tends to expose a bed that sits too high.
- Return to headphones and check that the music swells back cleanly at the end, without a sudden jump in level.
If your room is lively, check the voice track first. Room reflections and hiss make speech less distinct, and people then over-duck the music to compensate. Fixing the source with the tips in the guide to removing podcast background noise often lets the bed sit a few dB higher.
A practical step for this week: take your last episode's intro, convert whatever ducking is there to editable volume points, and nudge every downward fade earlier by 150 ms. Then do the phone speaker test. Most shows hear the difference immediately.
FAQ
How loud should background music be under podcast speech?
Aim for the music to sit roughly 20 to 24 dB below the voice while someone talks. Go lower for busy music or quiet speakers.
Is auto ducking good enough for a podcast?
It is a good first pass. Convert it to keyframes and fix the dips caused by breaths and laughter, and start each fade slightly before the first word.
Should music play under the whole episode?
Usually not. Beds help intros, transitions and outros. Under long conversation they add fatigue and make edits harder to hide.
What fade time sounds natural for ducking?
Around 150 to 300 ms going down and about one second coming back up works for most voices. Shorter returns tend to sound like pumping.
More in Podcasting
Podcasting
Podcast Cover Art Size and Design That Reads in Apps
Your show sounds good, but in the app it shows up as a blurry square with text nobody can read.
Podcasting
Podcast Mic Bleed and How to Stop It in a Shared Room
Two people at one table, two microphones, and every word from your co-host shows up a fraction of a second late in your own...
Podcasting
Podcast Episode Length and How Long Your Show Should Run
You finish recording, open the editor and see 74 minutes on the timeline. Is that too long?