One Voice Loud, One Voice Quiet. How to Match Podcast Levels
Your guest is half as loud as you and normalizing did not fix it. Match voices by loudness, not peaks, with per track steps that work in any editor.
By Don karlos Fru ยท Published 2026-11-01
QUICK ANSWERPut each voice on its own track, then match the tracks by loudness rather than by peaks. Raise the quiet track with gain until both voices measure close to the same loudness, aim the finished mix at about minus 16 LUFS for stereo or minus 19 LUFS for mono with true peak at or below minus 1 dB, and add light compression only at the end. If your host offers automatic leveling, that one setting does this whole job. On a single mixed track the fix is weak, because every change lifts or drops both voices together.
The question every interview show hits in the edit
A poster on the Audacity forum asked the plain version of it. How do you make two voice levels equal, with no voice higher or lower, just equal in volume? Another editor on the same forum had a recording with two voices and music, all at different loudness, and wanted them to sound the same. The Podcast Host runs a reader question column built on the mirror problem, a recording that comes back too quiet overall.
The mismatch is baked in at recording. Two people sit at different distances from two different microphones in two different rooms. One voice arrives strong and the other arrives thin, and the gap only becomes obvious when the headphones go on in the edit.
Editorial image from this article.Why normalizing by peaks leaves the guest quiet
Most editors reach for normalize first, and normalize works on peaks. Auphonic, the loudness company behind a lot of podcast processing, states the problem directly. Human loudness perception does not depend on peak levels, so recordings should be normalized by loudness rather than by peaks.
A quiet guest with one loud laugh can have the same peak as a loud host. Normalize both tracks to the same peak and the guest still sounds half as loud, because the average energy of the two voices never moved. The number that tracks what ears hear is loudness, measured in LUFS. Match that number between tracks and the voices sit together.
The target numbers for a podcast mix
Buzzsprout publishes the targets in its Magic Mastering help page. Aim for about minus 16 LUFS on a stereo mix and minus 19 LUFS on a mono mix, with true peak at or below minus 1 dB. The same page ties minus 16 LUFS, plus or minus 1 dB, to the authoring guidance Apple gives podcasters.
Those numbers are for the finished episode. The matching work happens one level down, track by track. Each voice should reach a similar loudness before the tracks are mixed, so no single voice carries the whole target alone.
The working order, step by step
Start with separate tracks. Every clean fix in the forum answers and in Descript mixing guidance rests on the same condition, one track per speaker. If your recording came back as one mixed file, work with gain envelopes on the quiet passages and accept a rougher result, then change the recording setup for the next episode so each voice lands on its own track.
On separate tracks, run loudness normalization across all tracks at once. In Audacity the effect is called Loudness Normalization, and the point of selecting every track together is that the voices get matched to each other, not just lifted one by one. One trap waits here. Audacity ships with a default of minus 23 LUFS, which is a broadcast default. Set the podcast target instead, minus 16 LUFS for a stereo show.
Then compress lightly, as polish rather than as the fix. A gentle compressor evens the peaks inside each voice, the laugh that jumps and the sentence that trails off. Descript mixing guidance gives the same shape. Keep each speaker on a separate track, aim for consistent loudness across all voices at around minus 16 LUFS for stereo, then use light compression or a cleanup tool such as Studio Sound to even out what is left.
Finish by listening on headphones at low volume. Level problems hide at high volume and show themselves when the playback is quiet, which is how a lot of listeners hear a show in a car or on a walk.
The one pass option inside your host
If the manual order sounds like a lot of steps per episode, the hosts sell the shortcut. Buzzsprout Magic Mastering includes a feature called Smart Adaptive Leveling, and the help page says it corrects the level differences between speakers and music and applies dynamic range compression for a balanced overall loudness. The algorithms are licensed from Auphonic. Auphonic itself sells the same loudness processing as a standalone service.
The honest trade is control. Automatic leveling gets a two voice interview to a good place in one pass. It cannot know that you wanted the guest a touch under the host for a serious segment, and it cannot repair a voice that clipped at recording. Clipping is a recording problem. No leveling tool restores a distorted peak.
Fix the next recording before it needs fixing
Most level gaps start as distance gaps. Ask your guest to stay a hand width from the microphone and to stop drifting back as the conversation warms up. Record a thirty second test with both voices at full talking volume and listen before the real session starts. Watch your own level as well. Hosts lean in and get louder as they get interested, and the guest stays where they started.
Remote recording tools that save one track per person make all of this easier, because the matching step assumes separate tracks exist. If your current setup mixes every voice into one file, that single fact explains most of your editing pain.
Even levels are only half of keeping a listener
A listener who can hear both voices without riding the volume button stays for the conversation. The next job is making sure that listener hears about the next episode. The episode itself is the best place to ask. Point listeners to your show page and invite them to join the email list for the notes and the next release.
That is the job PodSubs was built for. PodSubs gives podcasters a page where listeners join the show email list, so the audience you earn at the microphone stays reachable between episodes. Set it up once at podsubs.com and name the link at the end of every episode.
Frequently asked questions
Why does my guest still sound quiet after I normalize both tracks?
Normalize works on peaks, and ears judge loudness by average energy rather than by the single loudest moment. A quiet guest with one loud laugh can match your peak and still sound half as loud. Match the tracks by loudness in LUFS instead, which is the measure that tracks what listeners actually hear.
What loudness should a finished podcast episode hit?
Aim for about minus 16 LUFS on a stereo mix and minus 19 LUFS on a mono mix, with true peak at or below minus 1 dB. Buzzsprout publishes those targets in its Magic Mastering help page and ties the stereo figure to the authoring guidance Apple gives podcasters.
Can I fix uneven voices on a single mixed track?
Only roughly. On one mixed track every gain change moves both voices together, so the repair is a slow pass of manual gain envelopes on the quiet passages. Separate tracks per speaker are the condition for a clean fix, which is why recording tools that save one track per person matter.
What does Audacity get wrong by default for podcasts?
Audacity Loudness Normalization defaults to minus 23 LUFS, a broadcast setting. A podcast stereo mix should target about minus 16 LUFS. Run the effect across all tracks at once so the voices are matched to each other, then set the podcast target before you apply it.
Is automatic leveling from my host good enough?
For a standard two voice interview, usually yes. Buzzsprout Magic Mastering Smart Adaptive Leveling corrects level differences between speakers and music and applies compression for a balanced result, using algorithms licensed from Auphonic. It cannot repair a voice that clipped during recording, and it gives you less control than the manual order.