LUFS vs dBFS vs dBTP: What Your Audio Meter Means
LUFS describes overall loudness. dBFS describes saved sample levels. dBTP estimates peaks between those samples. You need both loudness and peak readings to judge an export.
By the fix.fm team · Published · Updated
Three measurements, three questions
A loudness meter and a peak meter answer different questions. If you raise a whole file by a fixed amount, both rise. If you compress only its loudest moments, the relationship changes. That is why a file with peaks close to zero can still sound much quieter than another file.
| Reading | The question it answers | The limit of the reading |
|---|---|---|
| Integrated LUFS | How loud does the whole programme measure? | It does not tell you whether every sentence is equally clear |
| Sample peak, dBFS | What is the highest saved sample? | It misses peaks between samples |
| True peak, dBTP | How high might the reconstructed waveform go? | A meter estimates it with interpolation; implementation matters |
LUFS uses frequency weighting and gates to keep very quiet material from dominating the result. The ITU-R BS.1770 recommendation describes programme loudness and true-peak measurement. The fix.fm meter uses the familiar mono/stereo K-weighted, gated approach and four-times interpolation. It is an estimate for checking recordings, not a certification for a broadcaster's delivery specification.
A tone experiment that separates the readings
The table below uses generated signals rather than a microphone recording. Every sample is defined by a formula, so you can recreate the experiment in a tone generator or a script. The numbers are produced by the same analysis code as our LUFS meter.
| One-second signal, 48 kHz | LUFS | dBFS peak | dBTP estimate |
|---|---|---|---|
| 1 kHz sine, one channel | -23.0 | -20.0 | -20.0 |
| Same sine, two identical channels | -20.0 | -20.0 | -20.0 |
| 12 kHz sine, 45° phase, amplitude 1 | 0.3 | -3.0 | -0.0 |
x[n] = amplitude × sin(2π × frequency × n / 48000 + phase). The first two rows use amplitude 0.1 and phase 0. The last uses amplitude 1 and phase π/4. The same meter code computes this table during the site build.The first row's sample peak is about -20 dBFS while its loudness is about -23 LUFS. That difference is a property of this sine wave and the weighting, not a conversion factor for every recording. Speech and music have different energy over time and frequency.
The second row copies the same signal to two channels. The loudness calculation sums both channels, adding about 3 LU, while the largest individual sample stays the same. Converting a mono file into duplicated stereo therefore changes the meter reading without creating new stereo detail.
The last row puts a high-frequency sine halfway between its largest possible saved samples. Its saved peaks sit below the waveform's actual maximum. A true-peak estimate finds a higher value. Real exports can have similar intersample peaks, so a sample ceiling alone is not a complete delivery check.
Choose the destination before the target
There is no universal loudness number for podcasts, music, broadcast and live calls. Spotify's artist guidance describes normal music playback at -14 LUFS, with different listener settings and device exceptions. That is a playback policy, not an instruction to make every spoken-word recording measure -14.
For podcast delivery, follow your host and destination. The podcast audio workflow links to Apple's current spoken-word guidance. Use those requirements for the final programme rather than forcing every isolated voice take to the final mix target.
Work out the gain, then check the ceiling
Suppose your file measures -23 LUFS and you want -16. A uniform gain change would be +7 dB. If its estimated true peak is already -3 dBTP, that change would push it to roughly +4 dBTP. The target cannot be reached with that simple gain alone while keeping peaks below -1.
You can keep a quieter target, manually reduce an unusually loud event, or use compression and limiting. Those decisions affect the sound, so listen afterward. The quiet-audio repair tool uses dynamic control when raising quiet speech; it is not merely multiplying every sample until it reaches a number.
Measure the saved file again
Do the final check on the delivery file, not just the editor's master. Lossy encoding can change peaks. Browser decoding may also resample the audio, and different interpolation filters can produce slightly different estimates. A small difference between two meters does not automatically mean one is broken.
Use a reference tone to check basic behavior, confirm mono or stereo layout, and compare like with like. For a contractual broadcast or mastering requirement, use the specified measurement tool and workflow. For everyday recording work, the free meter gives you a practical way to spot a quiet average, little peak headroom, or an accidental channel change.