YouTube Loudness Check

What this checks

• Integrated loudness (LUFS). The average loudness of the whole file, measured the way broadcast and streaming meters do it (ITU-R BS.1770-4). It's the number loudness normalization reacts to.
• True peak (dBTP). The highest point the waveform reaches, including between samples, estimated with 4x oversampling. Peaks right at 0 leave no room for YouTube's encoding.
• Long silences. Every stretch longer than 1.5 seconds where all channels stay below -50 dBFS, with the time it starts. It's the dead air worth a second look.
• When the sound starts. How long the file stays silent at the top before anything is heard.

What does YouTube do with loud and quiet videos?

In our 2026 loudness study, 50 recent long-form videos from 25 large channels measured anywhere from -24.3 to -7.4 LUFS, with a median of -16.3. Then I opened five of them on YouTube and read the player's "Stats for nerds" line:

• The three loud ones, measured at -7.4, -12.3 and -13.2 LUFS, all played at the -14 target. The extra loudness was simply taken away.
• The two quiet ones, measured at -23.9 and -24.3 LUFS, played at about -20: six decibels under the target. The player did not raise them.

So mixing louder than about -14 LUFS buys nothing, and mixing much quieter makes your video play softer than the one before it in someone's queue. That's why this tool reads your number against -14. The check was five videos in one desktop browser session; other devices and apps may behave differently. The full data, method and a downloadable table are in the study.

How the measurement works

1. Decoding. Your browser decodes the audio track and resamples it to 48 kHz, the same rate FFmpeg's EBU R128 meter works at.
2. K-weighting. Each channel goes through the two-stage filter from ITU-R BS.1770-4 (a high shelf that models the head, then a high-pass). The filter is computed for the sample rate, so it isn't tied to one fixed table.
3. Gating. The signal is cut into 400 ms blocks with 75% overlap. Blocks under -70 LUFS are dropped, then blocks more than 10 LU under the average of the rest are dropped too. Integrated loudness is the average of what's left. Surround channels are weighted 1.41 and the LFE channel is left out, as the standard says.
4. True peak. The loudest parts of each channel are oversampled 4x with a 49-tap windowed-sinc interpolator. This is an estimate: a mastering meter can differ by a few tenths of a decibel.
5. Silence. A stretch counts as silent while every sample on every channel stays below -50 dBFS; stretches of 1.5 seconds or more are listed.

Checked against FFmpeg. Before publishing, I ran this page in Chrome and FFmpeg's ebur128 filter on the same 16 real files: finished videos, audio downloaded from YouTube, WAV and MP3 exports and a raw camera clip, in AAC, Opus, MP3 and PCM, at 16, 44.1 and 48 kHz, from 28 seconds to 61 minutes long. Integrated loudness matched within 0.01 LU on 15 of the 16 files and within 0.11 LU on the last one, a 16 kHz mono track (FFmpeg itself reads that file 0.09 LU apart depending on whether it measures at 16 or 48 kHz). True peak matched within 0.12 dB. The silence check found the same 17 stretches as FFmpeg's silencedetect, with start and end times within 0.01 s.

Long video? Check just the audio

Your browser has to hold the whole decoded soundtrack in memory, which is about 23 MB per minute of stereo audio. Files over about 500 MB or 30 minutes can run out of memory, especially on a phone. The tool warns you before it starts and lets you try anyway.

The quick fix is to export only the audio from your editing app (WAV, or M4A/AAC for a smaller file) and check that. The loudness is the same as the video's, as long as you export the final mix.

FAQ