Problem with True Peaks (is it Limiter6 or Reason DAW causing issue?)

rob aylestone said:
But surely, level is a simple measurement of the level at one sample - so each sample, no matter what the rate is, has a value. If that value is every bit on, then that's 0dB? If you have a hardware machine with varispeed, turning the knob speeds up or slows down the pitch - but the volume stays exactly the same? It can't have a link to tempo as everyone would notice if a 120bmp track was louder or quieter than one at 90? Something very weird is happening here?
Click to expand...
He is talking about the daw time stretching the audio because he has it set up to match the audio to the tempo of the session. Since the daw is pitch shifting and resampling the audio, it can end up with peaks higher than the original audio.

It's best to turn that feature off, unless you are composing with loops and such.
 
True Peak has very little to do with actual sample values to begin with. It essentially makes up values of samples that would exist between the samples that are already there. Then also the sort of resynthesis done to stretch time without changing pitch has very little to do with the actual existing audio samples too. It’s basically trying to reconstruct an approximatlion of what it would sound like by combining a bunch of sine waves. Then any filter in the process will change the actual sample values and redistribute the energy of the wave that often can make actual sample peak values higher than they were before the filter. But basically if you measure a thing, then change that thing and measure it again, you really shouldn’t expect the two measurements to match because they’re not measuring the same thing.
 
ashcat_lt said:
Either it’s resampling to maintain the original pitch which is literally resynthesis so that it’s not even the same audio to begin with, or it’s playing back at a different speed and you’d hear a pitch shift because every frequency gets shifted. Either way, we’d expect the weird math that generates the completely imaginary and hypothetical TP numbers to work out different from when it was played at the correct original tempo.
Click to expand...


That doesn't make sense at all. Speeding up or slowing down a track doesn't change the volume.
 
I just created a drum pattern with a range of huge and small sounds in Cubase.

Drum loop created, placed on timeline. Copied twice. one had pitch raised 5 semi-tones, the other reduced in length to 50%. The loop them played in a cycled start/finish. The readings taken from the meter section.

I have no idea whatsoever happens to produce these results but pitch and speed change LUFS, and the RMS and Peak results. I can't work out what is happening - but the results are interesting.

The momentary LUFS is -15.8dB
The short term LUFS averaged is -21.5dB
The integrated LUFS is -23.5dB
The True Peak is -0.18dB
RMS max is -20.3dB
Peak Max is --4.9dB

Raising the pitch to 5 semi-tones above the original of the drum clip produces these result
The momentary LUFS is -16.6dB
The short term LUFS averaged is -22.3dB
The integrated LUFS is -24dB
The True Peak is -6.74dB
RMS max is -21.3dB
Peak Max is --6.7dB

Raising the tempo of the drum clip to double the original produces these result
The momentary LUFS is --16.6dB
The short term LUFS averaged is -22.3dB
The integrated LUFS is -24.4dB
The True Peak is --6.74dB
RMS max is -23dB
Peak Max is -8.5dB
 
True Peak produces a value representing the reconstructed waveform, to it could vary a bit with speed. But I wouldn't expect large variations.

[Edit] I now think TP levels could change substantially.
 
Last edited:
it may not change the peaks...
but the RMS will be greater in a faster song, because there is never any pauses in a fast constant mix....
 
True Peak isn't just measuring sample values, it's calculating peak values of the reconstructed waveform.
rob aylestone said:
But surely, level is a simple measurement of the level at one sample - so each sample, no matter what the rate is, has a value. If that value is every bit on, then that's 0dB? If you have a hardware machine with varispeed, turning the knob speeds up or slows down the pitch - but the volume stays exactly the same? It can't have a link to tempo as everyone would notice if a 120bmp track was louder or quieter than one at 90? Something very weird is happening here?
Click to expand...
True Peak isn't just reporting sample values, it's calculating peak values of the reconstructed analog waveform. If you increase the tempo, you make the slope of the wave steeper, which does seem to suggest that it could create a higher peak between two near-0 dBFS samples.
 
I'm not sure I quite get that? If you increase tempo, surely the waveform attack remains the same, and it's just longer between them. If the slope changed this would be like speeding up old tape systems where there was also a pitch change, and resultant timbral change. A snare drum pattern played faster would surely have the same sound for each thump? I can see that an averaging would result in lower readings when taken over a longer period (I think) a bit like dosimeters for people working with loud music. One trumpet blast might be a particular reading, level wise, but lots of them accumulate? Is this maybe what happens? I can't make any sense from my measurements at all.
 
I just did an experiment. I generated a 1 kHz tone at 0 dBFS in Sound Forge and imported it into Vegas Pro. Then I pushed the volume up 3 dB and rendered it to a new track. With the original track muted, the new track showed a TP value of +0.1. Then I time-compressed it (without preserving pitch) by different amounts (25%, 50%). The TP value went up to +0.2. The most interesting thing was that the TP value went up to +0.2 when I stretched it. When compressed by 75% (x4 playback speed), the TP went up to +0.3.
 
Gonna say this one more time.

If you have an audio recording of a snare drum pattern and you change it’s speed without changing the pitch, then you no longer have the same audio recording at all. The program did some form of analysis of the original signal and then synthesized a whole new audio recording that should sound more or less like the original might if it had been at the new speed, but the new waveform is actually created new from whole cloth and will not look anything like the original except by complete accident. It is almost guaranteed to measure different because it’s not at all the same.

bsg mostly covered the other case where its literally just playing the samples faster or slower, but also there is usually a filter involved in that sort of processes and filters can change peak levels due to phase smearing which redistributes energy and can in fact make peaks higher even if it is cutting some frequencies. I sometimes think of like the (lowpass) filter won’t let it “turn the corner” quite as quickly, so it ends up sometimes overshooting. This can cause a change to actual sample value peaks as well as imaginary “true” peaks.
 
You must log in or register to reply here.
Back
Top