Loudness Matching With Reference Tracks in FL Studio 2026: How to Align, What to Compare and the Measured Errors

A reference track is a finished piece a mix can be measured against. The comparison only works when both are equally loud: the louder piece sounds fuller and clearer no matter how it was mixed. That is why every comparison starts with loudness matching, and that is where the errors happen which this article puts numbers on.

The measurement used a purpose-built piece of 2:40 with an intro, two verses, two choruses and an outro, available in three versions: a mix without a limiter at −16 LUFS, a dense reference at −8 LUFS and a dynamic reference at −12 LUFS. Four questions were answered: what is left when matching happens by peak level? What does matching by loudness demand? How far off is a comparison that only uses the chorus? And what changes when the reference runs through the own master chain?

How this was measured

• The piece: 160 seconds at 120 BPM, 44.1 kHz, stereo, built synthetically from the same blocks as the measurement on true peak after encoding: an intro over 8 bars, a verse over 16, a chorus over 16, verse and chorus again, an outro over 8. Instrumentation and level change per section so that the piece has a dynamic profile close to a real case.
• Versions: the mix only turned up to −16 LUFS, without a limiter; the dense reference limited to −8 LUFS and −1 dBTP; the dynamic reference to −12 LUFS and −1 dBTP. Both references come from the same recording, so they differ only in processing.
• Tools: ffmpeg 7.1.5 on a Raspberry Pi 4. Loudness and loudness range according to ITU-R BS.1770 with the ebur128 filter, short-term loudness from the same run at one value per second, true peak computed separately with fourfold oversampling according to BS.1770-5.
• Third-octave bands: average power spectrum over windows of 16,384 samples, summed into third-octave bands from 25 Hz to 16 kHz. As a calibration the mix received a known shelf of +2 dB above 8 kHz; the measurement found it again as +1.94 dB at 10.2 kHz and +2.08 dB at 16.1 kHz.
• How far this carries: a synthetic piece is not a production, and a limiter in ffmpeg is not the limiter in FL Studio. The orders of magnitude still hold, because they follow from the definitions of the measures rather than from the taste of a plugin.

Why peak level fails as a yardstick

The three versions carry the same music but differ in everything that counts:

The dense reference is 8 dB louder than the mix and still carries the lower peak value. Bringing both to the same peak therefore does not shrink the difference, it grows it: after matching by sample peak, 10.78 dB of loudness difference remained, by true peak 9.25 dB. The EBU describes exactly this as the starting point of loudness normalisation in R 128, namely that peak normalisation of audio signals has led to considerable loudness differences between programmes, and that these inconsistencies are the cause of most viewer and listener complaints.

The second look goes to the loudness range: 12.4 LU for the mix, 0.5 LU for the dense reference. A span that small means every moment in that master is equally loud. That is not a detail, it is the very difference a comparison is supposed to reveal.

Matching by loudness, and what it demands

Only the last way reaches the goal, and it comes with a condition. To keep up with −8 LUFS the mix would have to reach a true peak of +8.2 dBTP; without a limiter that is not a file anything will play. From this follows the most practical rule in this article: for a comparison the reference is turned down, never the mix up. Lowering the reference by 8 dB costs nothing, raising the mix by 8 dB demands processing that distorts the comparison before it even starts.

How loud a finished master should end up is a different question; it is covered in the tutorial on mastering to streaming loudness. For the comparison itself it plays no part.

Which section gets compared matters

The chorus is the favourite place for a comparison. It is also the least representative place between a dynamic mix and a dense master, because there the two sit close together while in the verse they drift apart:

Matching on the chorus alone sets the mix 2.5 dB too quiet compared with matching over the whole piece; matching on the verse alone sets it 2.9 dB too loud. Both are more than most ears would still name as a loudness difference at that point, and that is exactly why R 128 requires the audio signal to be measured in its entirety, without emphasis on specific foreground elements such as speech, music or sound effects.

Against the dynamic reference the error was only 0.2 dB, because mix and reference follow a similar profile. The error therefore grows precisely when the difference is large, which makes it largest where it does the most harm.

How exact the match has to be

R 128 names two tolerances: for the implementation of loudness workflows, for example in quality control, ±0.2 LU is allowed to account for measurement errors; where the target level is not achievable practically, for example in live programmes, ±1.0 LU is permitted, and a deviation towards the limits of the tolerance should not become standard practice. For a comparison at the desk that means a match within 0.2 LU is both achievable and worthwhile, and anything beyond it is a setting that co-determines the result.

The reference track in FL Studio

The reference track goes into the playlist as an audio clip and gets a mixer track of its own; the mixer offers 500 insert tracks with ten effect slots each according to the manual. What matters is the path the signal takes from there: everything in the mixer is routed to the master track, unless an insert track is routed directly to an ASIO output. With an interface that offers more than one output, the reference can therefore bypass the own master chain.

Why that matters shows in the measurement. The dense reference was sent through a master chain of the usual kind, that is 3 dB of drive into a limiter with a ceiling of −1 dBFS:

Without drive the reference stayed unchanged, because its peak sat below the limiter ceiling. With 3 dB of drive it gained 1.6 LU, the dynamic one even 2.2 LU. A comparison against a reference treated that way measures the own master chain as well, in the same order of magnitude as the difference the comparison is actually about. Where no second output is available, the simple path remains: switch the master chain off while A/B-ing.

For metering inside the program there is Wave Candy in its peak meter mode. The metering algorithm can be switched there, among others to ITU-R BS.1770, the same frequency-weighted average over several seconds this measurement uses. For finished files the LUFS and true peak meter reads programme loudness, loudness range and true peak in one pass, without the file leaving the computer.

The level of the reference track is trimmed with the fader of its mixer track, not with an effect: an effect in the chain changes the signal, the fader only its level. And the trim value belongs in a note, because it holds for this one reference only.

What is left to compare once loudness matches

Once both are equally loud the actual comparison begins. The measurement shows how large the differences still are: after matching to the same programme loudness, mix and dense reference differed by up to 3.1 dB per third-octave band.

Both versions come from the same recording, and the mix was only turned up. What appears as a difference here was made entirely by the limiter: it pulled the highs above 8 kHz down by up to 3.1 dB and the region around 100 to 126 Hz by a good 2.5 dB. Seeing a deviation like that in a comparison and equalising the mix in response corrects an effect of loudness processing, not the mix. The loudness range says the same thing from the other side: 12.4 LU against 0.5 LU.

For the comparison itself this means equal loudness, the same place in the piece, short switching times, the same monitoring. Whatever still differs is a question of the mix, and there the chain from the tutorial on the vocal chain in FL Studio helps as much as a look at the sides with the tool for stereo correlation and mid/side.

What this covers and what it does not

• The piece is synthetic and identical in all three versions. Between two real productions the difference of the recording comes on top of the difference in processing; the third-octave values measured here are therefore a lower bound, not an upper one.
• The measurement used ffmpeg, not the plugins of FL Studio. A different limiter changes the spectrum differently.
• The tolerances from R 128 come from broadcasting. There is no equivalent specification for music production; they serve here as a yardstick for what counts as equally loud in measurement terms.
• The error from matching on the chorus depends on the piece. With a similar profile it was 0.2 dB, with a very different one 2.5 dB.
• The FL Studio details come from the Image-Line manual and were not measured.

Sources

• EBU R 128 (V5, November 2023): Loudness normalisation and permitted maximum level of audio signals: peak normalisation as the cause of considerable loudness differences, measurement of the signal in its entirety, tolerances of ±0.2 LU and ±1.0 LU, true peak not above −1 dBTP.
• ITU-R BS.1770: Algorithms to measure audio programme loudness and true-peak audio level: definition of LUFS and LU, gating and the fourfold oversampling for true peak.
• Image-Line: FL Studio Manual, Mixer: 500 insert tracks with ten effect slots each, routing to the master track unless an insert track goes directly to an ASIO output.
• Image-Line: Wave Candy: oscilloscope, spectrum, peak meter and vectorscope modes, metering algorithms peak, PPM, RMS, VU, Leq(A) and ITU-R BS.1770.
• Image-Line: Fruity Limiter: LIMIT and COMP tabs, controls of the limiter stage.