Does MSpectralDynamics do this with the sidechain signal?

[pumafred]

Was just reading a description of Trackspacer, and wondered if MSD treats the sidechain the same way:

Multiband compressors split the signal in different bands and apply different settings on each band, but the detection is still the same as regular ones. Trackspacer splits the source signal into 32 bands and analyses them individually. Then, it reverses the result and applies a 32 band EQ to duck the tracks on problem tracks by exactly the same frequencies, giving the source track the space it needs to occupy in the mix… and all in real time!

[DarkStar]

When compared to other frequency-dependent dynamics tools, like multiband compression, spectral shaping can offer far more resolution across the spectrum. By analyzing the signal across 32 Mel-spaced ("a perceptual scale of pitches judged by listeners to be equal in distance from one another") frequency bands, each band can be processed uniquely and without applying crossovers. You could imagine spectral shaping as a 32-band dynamic equalizer, with individual band shelf filters for every band, each automatically setting thresholds, time constants, and reduction amounts based on tuned models for each sound source.

https://www.izotope.com/en/community/bl ... aping.html

MSD may well take the same approach. If so, then I would expect that the analysed signal (either main or side-chain) is split into 32.

MSpectralDynamics uses a different approach as it works with the entire audio spectrum instead of bands. It approximates the energy located in each frequency and its surroundings and applies the dynamics to that frequency accordingly.

But I could be completely wrong

[jmg8]

Yes they can work the same. Spectral is a lot more accurate as it uses hundreds of bands instead of just 32.
It can also be done in MAutoDynamicEQ.

[pumafred]

Thanks, but neither in Izotope nor in MSD Info do I see clearly indicated if the sidechain signal is also split up. It IS clear that the signal to be processed is, and that MSD is much more powerful.

[jmg8]

Yes it is split up.
If it wasn't every band would detect the same thing and therefore behave exactly the same.

[DarkStar]

Yes they can work the same. Spectral is a lot more accurate as it uses hundreds of bands instead of just 32.
It can also be done in MAutoDynamicEQ.

Hundreds? I did not know that; got a link?

[jmg8]

I use the term bands loosely. But it is hundreds of bands because it is spectral. The frequencies are processed independently and there are thousands of frequencies.

[Gregorius]

I use the term bands loosely. But it is hundreds of bands because it is spectral. The frequencies are processed independently and there are thousands of frequencies.

Depends on your fft size/num filterbanks. And more bands does not always mean better quality (see a vocoder for example) or accuracy (since the resolution in time decreases).

[MeldaProduction]

Correct. However what spectral processing gives you is sort of addaptive bands concept - with vocoder (or any static multiband approach) you have predefined bands. Now imagine crossover points at 500 Hz and 2kHz, while at the same time having the fundamental at 500Hz (and harmonics at multiplies including 2kHz). With multiband processing you are sort of f**ed, because these signals just get into 2 neighbouring bands with the same level. Spectral processing on the other hand lets you use the Smoothness to improve the quality by sort of creating artificial bands that adapt to the signal, in our case you could think of a band that's covering 500 Hz and surroundings etc. But the price you pay is latency and preringing caused by linear-phase processing.