播客、YouTube、Spotify 与有声书的 LUFS 目标值（2026）

What LUFS should a podcast be?

Master a stereo podcast to -16 LUFS integrated with true peak no higher than -1 dBTP. That is the number Apple publishes ("around -16 dB LKFS, with a +/- 1 dB tolerance," true peak not exceeding -1 dB FS), and it travels well: Spotify lifts it about 2 dB on playback, YouTube leaves it alone, and it sits close to the -18 LUFS the Audio Engineering Society recommends for speech in AES TD1008.

Why not just use -14 like music? You can, and plenty of shows do. The cost is headroom. Getting speech to -14 LUFS usually takes more compression and limiting, and Spotify itself says masters louder than -14 should keep true peak below -2 dBTP because they distort more when encoded. For dialogue, -16 is the better trade. If you publish a mono file, aim for about -19 LUFS; the section on mono below explains the 3 dB.

What are the loudness targets for each platform in 2026?

"Target" means different things on different platforms. Some publish a delivery recommendation, some normalize on playback, and one publishes nothing at all. The last column says which.

One thing the table makes obvious: Spotify for Creators does not publish a loudness target for podcast episodes. The -14 figure comes from Spotify's music documentation, and the only podcast-specific Spotify number is the -16 LUFS ad spec. That is another reason -16 LUFS is a comfortable default for spoken word.

What do Spotify and YouTube actually do to your audio?

Spotify measures loudness on upload and adjusts the volume at playback; it does not rewrite your file. Louder masters are turned down to -14 LUFS. Quieter masters are turned up, but only as far as their peaks allow: Spotify's own example is a -20 LUFS track peaking at -5 dBFS, which only gets lifted to -16. On the Loud setting (-11 LUFS) a limiter engages at -1 dB. The web player and some third-party devices do not normalize at all.

YouTube only turns things down. A video measured at -9 LUFS plays about 5 dB quieter; a video at -20 LUFS plays at -20, which is why quiet tutorials sound weak next to everything else. You can check what happened to any video: right-click the player, open Stats for nerds, and read the Volume / Normalized line. The second percentage (or the "content loudness" figure in dB on older layouts) shows how much YouTube reduced it. A negative content loudness means your audio was louder than the reference.

What is the difference between integrated, short-term and momentary loudness?

All three come from the same ITU-R BS.1770 measurement; they differ in how long a window they average over. EBU Tech 3341 standardises the meter modes.

• Momentary uses a 400 ms window. It jumps around with every syllable and is mostly useful for spotting a sudden shout.
• Short-term uses a 3-second window. It is the one to watch while you level-match a host against a guest or an ad against the episode.
• Integrated is the whole programme, with gating: blocks below -70 LUFS are ignored, then anything more than 10 LU below the running average is ignored too. Gating is why long silences do not drag the number down. Every platform target in the table is an integrated value.

Loudness range (LRA, defined in EBU Tech 3342) tells you how much the short-term loudness moves. Conversational podcasts usually sit comfortably in single digits. A very wide LRA means listeners in cars and on trains will be reaching for the volume, and gentle compression before normalizing helps more than a lower target does. Our guide on normalization vs compression covers the order.

Why does true peak matter more than sample peak?

A sample-peak meter reports the highest stored sample. The waveform that a playback device reconstructs between those samples can go higher, and lossy encoders (MP3, AAC, Ogg Vorbis) push those in-between peaks higher still. BS.1770 defines true peak by oversampling the signal (at least 4x at 48 kHz) to estimate those overshoots, reported in dBTP.

In practice, a file whose samples never exceed -1.0 dBFS can measure around -0.5 dBTP or even above 0 after encoding if it is heavily limited. Apple asks for -1 dBFS true peak, Spotify for -1 dBTP (-2 if louder than -14 LUFS), and Apple notes that encoders "might clip the signal if the recommended true-peak value is not respected." Use a true-peak limiter, or leave an extra half decibel when your limiter only watches sample peaks.

Should a mono podcast be -19 LUFS?

Usually, yes. BS.1770 sums the power of each channel. Take a mono voice and put the same signal in both channels of a stereo file, and the stereo file measures about 3 LU louder, even though it sounds the same on one speaker. Most players send a mono file to both ears or both speakers, so a mono file at -19 LUFS sounds roughly as loud as a stereo file at -16 LUFS.

Apple's page does not mention mono at all, so -19 LUFS mono is common practice rather than an Apple rule. It is a good rule, though. If you are unsure, export stereo at -16 LUFS: dual-mono stereo costs bitrate, not quality, and removes the guesswork. The mono vs stereo guide covers when each makes sense.

Do audiobooks use LUFS?

Not for ACX. Audible's production platform requires every file to measure between -23 and -18 dB RMS, with peaks no higher than -3 dB and a noise floor no higher than -60 dB RMS, delivered as 192 kbps or higher constant-bit-rate MP3 at 44.1 kHz, one chapter per file, each under 120 minutes. RMS is not K-weighted and not gated, so a LUFS reading is only a rough guide; for narration the two numbers often land close, but close does not pass a check.

A practical approach: aim for about -20 dB RMS, which leaves room on both sides, keep true peak at or below -3 dBTP so MP3 encoding does not push peaks over, and fix the noise floor at the recording stage, because raising the level to meet the RMS window also raises the noise. ACX's own Audio Lab checks RMS, peak and format, but its help page says it does not detect noise floor problems, so check that yourself.

How do you check loudness and hit a target?

1. Finish the edit first. Cut silence, remove noise and level the speakers before measuring. Integrated loudness changes when you remove dead air, because gating only ignores the very quiet parts.
2. Measure. Load the finished file into the LUFS meter to read integrated loudness, short-term and momentary maximums and true peak, all calculated in your browser.
3. Normalize. The audio normalizer has one-click targets of -14 LUFS (YouTube and streaming), -16 LUFS (podcasts) and -23 LUFS (EBU R128). It measures integrated loudness with BS.1770 K-weighting and gating, applies the gain, and runs a peak limiter with a -1 dB ceiling. The limiter watches sample peaks, so read the true-peak result it shows afterwards; if it reads above -1 dBTP, normalize half a decibel lower. There is no -19 LUFS mono or ACX RMS preset, so for mono, normalize to -16 and lower the gain by 3 dB in your editor.
4. Export losslessly, then encode once. Deliver WAV to your host when it accepts it, or encode the MP3 yourself from the normalized WAV and measure the MP3 again.

Common loudness mistakes

• Peak normalizing to 0 dBFS and calling it done. Peak normalization ignores perceived loudness, and 0 dBFS leaves no room for encoder overshoot.
• Mastering speech to -10 LUFS "so it sounds loud". Spotify and YouTube turn it down anyway, and what remains is the squashed sound of heavy limiting.
• Leaving YouTube voiceover at -22 LUFS. YouTube will not turn it up. Your video simply plays quieter than the one before it.
• Treating ACX RMS as LUFS. Measure RMS for audiobooks, and remember the -3 dB peak limit, which is stricter than any podcast spec.
• Normalizing before editing. Removing a few minutes of silence or a loud outtake changes the integrated value. Normalize last.
• Comparing a mono file against a stereo target. A mono file reading -19 LUFS is not 3 dB too quiet.

For the signal chain around normalization, see how to normalize spoken-word audio and the podcast editing checklist.