LUFS vs dBFS: Can You Convert LUFS to dB?

"Convert LUFS to dB" is one of the most common audio questions, and the honest answer is that you can't, at least not with a formula. dBFS and LUFS both use the digital full scale as their reference, but they measure different things. This article explains the difference and what to do instead.

1. Two Different Measurements

dBFS describes the level of samples relative to the largest value the format can store. A sample peak meter reads the highest absolute sample, and an RMS meter averages the energy over a short window. Neither knows anything about hearing.

LUFS (Loudness Units relative to Full Scale, called LKFS in ITU and ATSC documents) is defined by ITU-R BS.1770. The meter:

1. applies a K-weighting filter: a high-shelf boost of about 4 dB at high frequencies and a high-pass that discounts low frequencies;
2. computes the mean square of each channel and sums the channels, with extra weight on surround channels;
3. for integrated loudness, gates out silence and quiet passages: blocks below −70 LUFS, then blocks more than 10 LU below the loudness of the blocks that remain.

EBU Tech 3341 adds the time windows that loudness meters display: momentary loudness (400 ms window), short-term loudness (3 s window) and integrated loudness over the whole programme.

2. Why There Is No Fixed Conversion

Two files with the same peak level in dBFS can have very different integrated loudness, because loudness follows the average energy of the programme, not its highest sample. The gap between the peak and the loudness is the peak-to-loudness ratio (PLR), and it depends on the content, not on a constant.

The frequency weighting adds a second dependence: a bass-heavy signal reads lower in LUFS than a mid-range signal with the same RMS level in dBFS, because the K-weighting filter discounts low frequencies.

3. The One Case Where It Works

BS.1770 is calibrated on a test signal: a 1 kHz sine at 0 dBFS in one front channel reads −3.01 LUFS. The −3.01 comes from the RMS of a sine, which is 3 dB below its peak. The same sine at 0 dBFS in both channels of a stereo file reads about 0 LUFS, since the two channels add their energy.

For a steady 1 kHz tone, LUFS therefore follows the dBFS level exactly. Use this to check that a meter is set up correctly, not to estimate the loudness of real programme material.

4. LU Differences Are Decibels

What does convert directly is a difference. One loudness unit (LU) is one decibel, so loudness normalisation is a simple gain: a track measured at −20 LUFS needs +6 dB to reach −14 LUFS. The catch is the peak level: if that track already peaks at −3 dBFS, adding 6 dB pushes it over full scale and you need a limiter, or a lower target.

5. Measuring Loudness in Practice

FFmpeg ships a BS.1770 meter. The summary at the end gives integrated loudness, loudness range and true peak:

ffmpeg -i input.wav -af ebur128=peak=true -f null -

In Python, pyloudnorm measures and normalises in a few lines:

import soundfile as sf
import pyloudnorm as pyln

data, rate = sf.read("input.wav")
meter = pyln.Meter(rate)                 # BS.1770 meter
loudness = meter.integrated_loudness(data)
print(f"{loudness:.1f} LUFS")

normalized = pyln.normalize.loudness(data, loudness, -14.0)

pyloudnorm applies the gain only: check the peaks afterwards, as explained in the previous section.

6. Loudness Targets

The table lists only targets published by the standards body or the platform itself (EBU R128, ATSC A/85, Spotify for Artists, Apple Podcasts audio requirements). Platform values can change, so check the current documentation before mastering. dBTP (true peak) is measured on an oversampled signal: it catches the peaks between samples that a sample peak meter in dBFS misses, which is why the ceiling is set below 0.

For recording levels in dBFS, see −18, −12 and −20 dBFS reference levels.