24 vs 6 dB, are there different dB in Live?

[black.onyx]

I try to get an understanding of the loudness in Live. I know there is differences between dB in acoustics and dB in signal processing. In signal processing there are even differences between power and amplitude AI tells me.

I would expect large warning signs everywhere pointing out the differences. Instead, I encounter long-time music producers discussing dB and completely talking past each other.

I produce white noise, run it through the "Autofilter" with a slope of 24 dB per octave, and examine the result in the "EQ Eight". I measure a drop of 6 dB from 1 kHz to 2 kHz, thus indicating a slope of 6 dB per octave.

24 vs 6, that's a factor of four. Even within Live, the dB apparently have different meanings. Can someone explain this in more detail?

[yur2die4]

I’m not even remotely an expert. Filters have various behaviors and consequences due to their architecture. I’m not sure how a resonant peak / Q might be related to factoring things. But

A big standout to me is that you are relying on a visual UI for doing your measurements. Visual interfaces also have their quirks and are not always perfect or reliable. Though there are some out there which are probably at least much more reliable than others.

[black.onyx]

Measured in dB, the slope is linear in EQ Eight. I can draw a line and determine the slope accurately enough. The deviation by a factor of 4 cannot be explained by inaccuracies. It's quite certainly a different scale.

The Live Autofilter and the filter of the Jun6V from Arturia deliver similar results here. For filters, there seems to be agreement on the dB unit even across company boundaries.

There must therefore be a certain standard. However, I can't find any reliable information. For classic search engines, the search terms are too general. AI fails with technical questions when it gets interesting.

[benmuetsch]

I'm not entirely sure what you are referencing to but the 24dB/Oct filters in filters are 24dB/Oct filters, also in Ableton Live.

That also has nothing to do with "loudness". In the analogue domain there are various filter topologies, mostly varying in Q and feedback (to put it simply).

If you use EQ8 for example, the standard Low-/Highpass is 12dB/oct, if you now use a Q of .71 you'll get -3dB at the cutoff point, if you use .5 Q, you'll have -6dB (representing roughly Butterworth and Linkwitz-Riley filters and their phase shift respectively).

[benmuetsch]

Just verified: set an Autofilter, Clean filter mode, 24dB HPF, 1kHz. Put a sine wave through, let's say 1200Hz, then compare to 600Hz, it's exactly 24dB down.

[black.onyx]

I don't use any Q settings of EQ Eight at all. I only use it as a display of the slope of the white noise.

Let's repeat your settings. I create sinus of 600 and 1200 by playing notes D with Analoge. I set the loudness to -12 dB in the mixer (similar in Youlean loudness meter by triggering true peak max repeatedly). Both D are similar here.

When I turn on the Autofilter at 1 kHz it affects both ends. The upper D gets -19 dB, the lower D gets -35 dB. That is at least 16 dB per octave. A strong difference to the slope in EQ Eight for white noise.

[x3000]

The unit dB only indicates a ratio between 2 values. However, this ratio is (decadic) logarithmic.
1 Bel (B) = 10 dB
Value 1² / value 2² = log(value in Bel)

e.g:
log(1²/4²) = -1.2 B = -12dB

If the control in the mixer is set to -12dB, this attenuates the level to 1/4 of the level before the mixer.


With the filter:
An octave up represents a ratio of 2:1.

The dB/octave only represents the slope of the filter. If the slope is very steep (high dB value), the curve drops quickly. If the slope is rather flat (low dB value), the curve falls more slowly.

A low-pass filter set to 1000 Hz only takes effect from 1000 Hz. If I send a white noise through this low-pass filter (24dB), I can detect a reduction of -24dB at 2000 Hz after the filter.

It is also not possible to read in which mode you are operating the filter: LP, HP, BP?

What does your mixer at -12dB have to do with it?


You are sending 600 and 1200 Hz through the filter, whose frequency is set to 1kHz (lowpass?) with a slope of 24dB. But the filter only filters from 1000Hz.

[benmuetsch]

I don't use any Q settings of EQ Eight at all. I only use it as a display of the slope of the white noise.

Let's repeat your settings. I create sinus of 600 and 1200 by playing notes D with Analoge. I set the loudness to -12 dB in the mixer (similar in Youlean loudness meter by triggering true peak max repeatedly). Both D are similar here.

When I turn on the Autofilter at 1 kHz it affects both ends. The upper D gets -19 dB, the lower D gets -35 dB. That is at least 16 dB per octave. A strong difference to the slope in EQ Eight for white noise.

Are you maybe using complex waveforms with Analoge?

[black.onyx]

@x3000:

My original test was running white noise through a 24 dB/octave filter. The behaviour doesn't differ much between LPF and HPF. I took my values from points within the spectrum of the EQ Eight.

BenMuetsch tetst is different. He uses sinus. That is a single wave. So you can measure it with the usual tools to measure the overall loudness.

Autofilter, Clean filter mode, 24dB HPF, 1kHz. Put a sine wave through, let's say 1200Hz, then compare to 600Hz, it's exactly 24dB down.

My second setup was a replication of his test. I get 16 dB down. Hence, we are getting much closer. Again, I don't observe a relevant difference in the steepness of LPF or HPF slopes. It's just the expected inversion.

[black.onyx]

@benmuetsch:

I suspect the cause of the difference is, that white noise is a large number of waves. Your result gives a 24 dB per octave slope for just one single sine wave. Obviously your approach is how the steepness of the filter is defined.

In practice the filter has to work on a many overtones. That is more in the direction of white noise. How this multitude affects the drop ratio is still a little too complex for my current level. At least the ratio seems to get smaller for any point within white noise.

[x3000]

How did you create the white noise?

The white noise that you can generate in Live produces a crooked curve in the spectrum analyzer (blank, without filter).

You need a technically reasonably clean white noise with the same level over the entire frequency range.

[black.onyx]

The white noise that you can generate in Live produces a crooked curve in the spectrum analyze

Confirmed. If I use the noise~ object Max for Live I observer an increase of 4dB between 100 Hz and 20 kHz. Within the range of 600 Hz and 1200 Hz it is negligible. For Jun6V it drops above 10kHz.

[x3000]

A picture is more meaningful:

https://ibb.co/NNYkbDb
pure white noise


https://ibb.co/c3M17G8
white noise with filter

[benmuetsch]

Noise or other complex waveforms consist - as the name says - of complex ovetone structures so don't expect exact subtraction of 24dB/Oct in the sum of the level (what you see on a fader/meter).

The filter has a passband, a stopband and to be picky - a transitionband. Complex waveforms always contain frequencies which will be present in all bands, so an exact reduction in amplitude is frequency and energy dependant.

[Synthbuilder]

The dB/octave figure is the eventual roll off of the filter. It is the slope of the filter at a frequency well above (for low pass) and below (for high pass) the cut-off frequency*.

For us audio folks the eventual roll off is often not as important as what it is doing around the cut-off point. The slope near the cut off point is not the final slope. The type of filter and the resonance (Q) setting will often have more of an affect on the sound than the eventual roll off. Indeed, a -24dB/octave low pass filter will increase the signal level around its cut-off point if the resonance is high.

* More complex filter types, such as elliptic filters, can produce roll off curves that are non trivial. The roll off is then described over a specific frequency band because its final roll off doesn't tell us much about its behaviour around the cut-off point.