What LUFS should a podcast or voice recording hit, and how do you get there?

Apple asks for -16 LUFS with a -1 dB true peak, and Spotify's -14 is a music number. What LUFS measures, how it relates to dB and RMS, what normalizing really does, and free ways to hit the target on a Mac.


You finish editing a forty-minute episode in Audacity. The waveform looks healthy, the loudest laugh peaks at -0.4 dB, so you run Normalize to -1 dB and export. Then you play it in a podcast app right after another show and have to reach for the volume button: your episode is clearly quieter. Nothing was wrong with the peaks. The problem is that peaks say almost nothing about how loud speech feels, and the app you are listening in does not care about peaks at all. It cares about LUFS.

This page answers two things: which LUFS number a podcast or a voice recording should hit, and how to get there with free tools on a Mac. It also covers the confusing neighbours of that question: what normalizing audio actually does, how LUFS relates to dB and RMS, and why a steady voice at the microphone saves more work than any loudness setting at the end.

One quick redirect first. If your real problem is that the recording clips or crackles at the loudest moments, the fix happens before LUFS ever enters the picture, and gain staging is the page you want. If people on calls tell you that you are too quiet, that is a live input problem, and why your mic is so quiet covers it. LUFS is about finished files.

The short answer: -16 LUFS integrated, true peak at -1 dB

For a spoken-word podcast, aim for -16 LUFS integrated loudness with true peaks no higher than -1 dBTP. That is Apple's published target, and the same -16 appears in Spotify's spec for podcast ads, which is stricter on peaks (-2 dBTP).

Here are the published numbers as of October 2026, taken from each organization's own pages:

Where the file goesIntegrated loudnessPeak ceilingNotes
Apple Podcasts-16 LKFS, ±1 dB-1 dBFS true peakMeasured per ITU-R BS.1770
Spotify, music (Normal setting)-14 LUFSbelow -1 dBTPPremium users can also pick Loud (-11) or Quiet (-19)
Spotify, podcast ads-16 LUFS, ±1.5-2 dBTPSpotify's ad spec, a good proxy for speech
European TV and radio (EBU R 128)-23 LUFS, ±0.5 LU-1 dBTPBroadcast delivery
US TV (ATSC A/85)-24 LKFSnot covered hereReference level for broadcast
Audible audiobooks (ACX)-23 to -18 dB RMS-3 dB peakNot LUFS at all, plus a -60 dB RMS noise floor

Two things jump out. First, the streaming numbers cluster between -14 and -16, while broadcast sits much lower, at -23 and -24. Broadcast is built for a TV in a quiet living room with volume to spare. Podcasts get played on a phone on a noisy commute, so they are mastered hotter. Second, Audible does not use LUFS at all. If you narrate audiobooks, you are working to an RMS window, which is a different measurement, explained below.

Why -16 and not Spotify's -14? The -14 on Spotify's loudness page is written for music. Pushing speech to -14 usually means heavier limiting to keep the peaks under the ceiling, and Spotify itself asks for -16 in its podcast ad spec. If you publish to every app from one file, -16 is the safer single number.

What LUFS actually measures

LUFS stands for Loudness Units relative to Full Scale. It comes from the ITU-R BS.1770 standard, and LKFS (the name Apple and US broadcasters use) is the same unit under an older label. When Apple says -16 LKFS and an editor says -16 LUFS, they mean the same thing.

Three ideas make it different from a plain level meter:

  1. K-weighting. Before measuring, the signal passes through a filter that roughly follows how ears hear: low rumble counts for less, the upper midrange and treble count for more. A bass-heavy hum and a bright voice at the same electrical level do not score the same.
  2. Gating. Silence is ignored. The integrated measurement drops anything below -70 LUFS outright, then drops anything more than 10 LU below the level measured after that first cut. Long pauses between sentences do not drag the number down, which matters a lot for speech.
  3. Time windows. A LUFS meter typically shows several readings at once:
ReadingWindowWhat it tells you
Momentary (M)400 msHow loud this syllable is
Short-term (S)3 sHow loud this sentence is
Integrated (I)the whole fileThe number platforms compare against their target
Loudness range (LRA)the whole fileHow much the loudness swings between quiet and loud passages

When a platform says "-16 LUFS", it means the integrated value. Your momentary reading will jump by several units from one syllable to the next during normal speech, and that is fine.

LUFS to dB: how the numbers relate

A common search is "LUFS to dB", as if there were a formula. There is a simple part and a part with no formula at all.

The simple part: one loudness unit equals one decibel of change. If a file measures -20 LUFS and you want -16, you add 4 dB of gain. Turn it down 3 dB and it measures 3 LU quieter. Gain changes translate one to one.

The part with no formula: LUFS and peak dBFS describe different things, so you cannot convert one into the other. Two files can both peak at -1 dB and measure -12 and -24 LUFS. A dense, compressed voice has peaks close to its average; an uncompressed voice with one big laugh has a peak far above everything else.

That gap is where people get stuck. A worked case:

Measured file:   integrated -21.0 LUFS, true peak -2.5 dBTP
Target:          integrated -16.0 LUFS, true peak -1.0 dBTP
Gain needed:     +5.0 dB
Peak after gain: -2.5 + 5.0 = +2.5 dBTP  (over full scale, will clip)
Over the limit:  +2.5 - (-1.0) = 3.5 dB that a limiter must catch

Plain gain cannot do this job. Either a limiter shaves 3.5 dB off the loudest moments, or a compressor evens the voice out first so that the peaks sit closer to the average. That is the real reason LUFS and compression get mentioned together.

LUFS vs RMS

RMS is the older way to express average level: it squares the signal, averages it and takes the root. It has no ear-shaped filter and no gate. Two consequences for voice:

  • Pauses count. A slow narrator with long breaths reads lower in RMS than a fast talker at the same perceived loudness, because the silence is averaged in.
  • Low end counts fully. Room rumble or a proximity-effect boom adds to RMS even though listeners barely register it as loudness.

LUFS fixes both, which is why streaming platforms publish their targets in LUFS. RMS still matters in one place a voice person will meet: Audible's ACX submission rules, which ask for -23 to -18 dB RMS, peaks below -3 dB and a noise floor under -60 dB RMS. If you record audiobooks, measure RMS for ACX and LUFS for everything else, and do not try to convert one into the other.

What normalizing audio does, and what it does not

"Normalize audio" means two different operations, and audio editors do not always say which one they run.

Peak normalization finds the single loudest sample and turns the whole file up or down until that sample hits your chosen level, say -1 dB. That is what the classic Normalize command does. It guarantees nothing about loudness: one cough can decide the gain for forty minutes.

Loudness normalization measures integrated LUFS and turns the whole file up or down until it hits your target, say -16 LUFS. Audacity has this as a separate effect: Effect > Volume and Compression > Loudness Normalization. Its default is -23 LUFS, the broadcast number, so change it to -16 for a podcast.

What both kinds have in common: they are a single gain change. The quiet parts and the loud parts move by exactly the same amount. Normalizing does not even out a voice that drops off at the end of sentences, does not tame a shout, does not remove noise, and does not change tone. If the recording had a 20 dB gap between your guest and you, it still has a 20 dB gap after normalization, just at a different overall level.

Streaming apps do the same thing on playback. Spotify's loudness page describes adjusting tracks toward the level the listener picked, with a limiter only on the Loud setting. Your file is not re-mixed; it is re-leveled.

Normalization vs compression

Compression changes the shape of the level over time: when the voice gets louder than a threshold, the compressor turns it down by a ratio, so quiet and loud passages move closer together. Normalization moves the whole file. A practical order for a voice track:

1. Clean up      high-pass around 80 Hz, noise reduction if needed
2. Even it out   compressor on the voice (gentle ratio, a few dB of gain reduction)
3. Set loudness  loudness normalization to -16 LUFS integrated
4. Catch peaks   true-peak limiter with the ceiling at -1 dBTP
5. Measure       check I and TP on the exported file, not inside the project

Normalization comes after compression because compression changes the integrated loudness. Normalize first and compress afterwards, and you are measuring a file that no longer exists. The limiter comes last because it only has to catch the few peaks the gain change pushed too high.

How much compression a voice needs, and how to set threshold and ratio without squashing it, belongs to a separate topic: how EQ and compression make a voice sound better walks through both.

A free LUFS meter on a Mac

You do not need to buy anything to measure LUFS. As of October 2026:

  • Youlean Loudness Meter 2 (free version). A standalone app and a plugin (AU, VST3, AAX) for macOS and Windows. The free version shows integrated loudness, true peak, loudness range and more, and does not expire. Drop your exported file into the standalone app and read I and TP. The paid PRO version adds ready-made presets for Spotify and Apple Podcasts, which you can replace with two numbers written on a sticky note.
  • Audacity. Measures and fixes in one place through the Loudness Normalization effect. Watch the "Treat mono as dual-mono" box, which is on by default; more on that below.
  • Logic Pro's Loudness Meter. If you already own Logic, its Loudness Meter plug-in shows momentary, short-term and integrated loudness.
  • ffmpeg's loudnorm filter. A free command-line route that measures and normalizes, covered in the next section.

Searches for "LUFS online" cover two different needs. The Youlean Online Loudness Meter is a free browser page that measures what your microphone hears right now, processed locally, nothing uploaded; it shows momentary and short-term LUFS, which is handy for checking speaking level before a recording. To normalize a finished file online, a web service such as Auphonic processes uploaded audio and has a free allowance of 2 hours per month as of October 2026. If you are weighing it against Podcastle, Auphonic or Podcastle compares the two.

The mono trap

Many voice podcasts are mono. Some meters measure a mono file 3 LU lower than the same audio stored as two identical channels, because stereo playback sends that one channel to both speakers. Audacity compensates by default ("Treat mono as dual-mono"); ffmpeg's loudnorm does not unless you set dual_mono=true. The same mono file can therefore read -16 in one tool and -19 in another. Pick one tool, check how it treats mono, and stick with it.

Normalizing with ffmpeg in two passes

ffmpeg ships with a filter called loudnorm built on EBU R 128. Its defaults are broadcast values (-24 LUFS, -2 dBTP), so set the targets yourself. The first pass only measures:

ffmpeg -i episode.wav -af loudnorm=I=-16:TP=-1:LRA=11:print_format=json -f null -

It prints input_i, input_tp, input_lra and input_thresh. Feed those into the second pass (the numbers below are example values from a first pass):

ffmpeg -i episode.wav -af loudnorm=I=-16:TP=-1:LRA=11:measured_I=-21.0:measured_TP=-2.5:measured_LRA=6.2:measured_thresh=-31.4:linear=true -ar 48000 episode-16lufs.wav

With all four measured values supplied, loudnorm tries to apply one linear gain change and leave the dynamics alone. The documentation sets two conditions for that: the target LRA must not be lower than the file's own LRA, and the gain must not push true peaks past the TP ceiling. Break either one and the filter reverts to its dynamic mode and starts shaping the level.

The example values break the second condition. They describe the same file as the worked case above: -21 LUFS with peaks at -2.5 dBTP, so +5 dB would put the peaks at +2.5 dBTP. That file gets dynamic processing; a file with more peak headroom gets the clean linear gain. The same logic explains LRA=11 instead of the default 7: if a conversation measures wider than 7 LU, a target of 7 would force dynamic mode for no good reason. One more detail from the documentation: in dynamic mode the filter upsamples to 192 kHz to detect true peaks, hence the explicit -ar 48000 on the output.

Measure the result again with the first command. If it lands within half a unit of -16, you are done.

Why the input decides how much of this you do

Loudness normalization is one gain change. That is its strength and its limit, because it lifts everything by the same amount, including the room.

Say you record a little quietly, the file measures -28 LUFS, and the background hum of the room sits at -62 dB. To reach -16 you add 12 dB. The hum now sits at -50 dB. If you were aiming at Audible's -60 dB noise floor, you just failed it at the export step, even though the original recording passed. The same happens with level swings: if your voice drifts 8 dB between leaning in and sitting back, normalization keeps that 8 dB, and the compressor has to work harder to hide it.

So the cheapest loudness fix is a steady, healthy level at the mic. That means the right distance, gain set so normal speech peaks well below 0 dB (the gain staging guide has the numbers), and a quiet room. Get those right and the chain above becomes a light touch instead of a rescue.

Where TunedMic fits, and where it does not

TunedMic is a Mac app that adds a virtual microphone tuned to your voice after a one-minute Sound Check. It processes the real mic live on the Mac, and any app that lets you pick a mic (Zoom, OBS, Audacity, a browser) can record from it. Among its manual controls are noise, room echo, clarity, loudness and a 12-band equaliser, and the nine profiles are built from your own Sound Check.

What that means for this topic: if you record into an app through TunedMic, the voice arrives with less room noise and echo under it, at a loudness you set once with the loudness control, so normalization has less noise to lift along with the voice. That is the honest scope. TunedMic has a loudness control, not a LUFS target: it does not measure integrated loudness and it does not promise that a file lands at -16. You still measure and normalize the exported file with one of the tools above.

Know its limits before you look further. It needs a Mac with Apple silicon (M1 or newer) and macOS 14 or later; there is no Windows or Intel version. It is most useful when the same voice problem shows up everywhere: in recordings, on calls, on streams. For a single file that is already recorded, a repair tool after the fact is the better match, and removing background noise from a recording covers that route.

Quick answers

Should I master my podcast to -14 for Spotify? You can, but -16 is the safer single number for speech. Spotify's -14 figure is its music reference; its own podcast ad spec asks for -16.

Does normalization lower quality? A gain change down does nothing harmful. A gain change up raises noise along with the voice, and a limiter catching too many peaks can flatten speech. Both are reasons to record at a sensible level, not reasons to skip normalizing.

Is -16 LUFS the same in every meter? For stereo files, yes, within a fraction of a unit. For mono files, check the dual-mono setting described above.

Do Zoom or Telegram calls use LUFS? No. A call has no finished file to measure; what the other side hears depends on your input level and the app's own processing. If calls are the problem, start with why your mic is so quiet.

What to do next

Export your latest episode as WAV, drop it into the free Youlean meter and write down two numbers: I and TP. If I is far below -16, check your recording gain before you touch any loudness setting. If I is close but TP is above -1, add a limiter and run the ffmpeg two-pass or Audacity's Loudness Normalization at -16. And if the voice going into the mic keeps fighting you, watch the 38-second video on the TunedMic home page to see the setup: you read aloud for a minute and the mic gets tuned to your voice on your Mac. The free 7-day trial includes every feature and needs no card or account, so you can try TunedMic on your own mic, and the price is on the pricing page.

Keep reading