ABOUT THE DRUMS SSX PLUGIN ANALYSIS WINDOW CONTROLS

Drums SSX is an acronym that stands for “Drum signal Source Fx Active
Separator and Re-mixer”. It's an all in one solution that is aimed
Use the blue active button on the upper left to engage or by-
at separating acoustic mono or stereo drum recordings into
pass the effect altogether.
their components so they can be remixed individually whenev-
er there is no multitrack recording available. This plugin also
NMF Mode
works great for extracting trigger signals.
Use this option to switch between two variants of the NMF al-
Drums SSX offers an advanced NMF-based live separation al- gorithm. The progressive mode yields stronger separation
gorithm including four simple preset kits, a comprehensive between the instruments but is prone to artifacts. Use the ad-
analyzer view, a simple to use and very musical sounding mix- aptive mode for more natural and blended sounding results
er that offers a 3-band linear phase EQ and dynamics section where desirable, especially in remixing applications. Depend-
per channel. Use the learn feature to further improve separa- ing on your input material the difference between the two can
tion quality and the im-/export kits matched to your setup. be rather subtle.

Inclined readers can learn more details in the “A Brief Essay on Separation
Drums SSX' core algorithm” section below.
The separation between instruments is achieved by means of

1
time variant Wiener filter masks that are constantly adapted to Im-/Export
the incoming signal (see system block diagram below for more
Click the im-/export buttons to load or save a set of custom
details). This setting essentially controls the dynamic of these
trained instruments to a file on your disk. This allows for ex-
filter masks. The weak option will be perceived as more natur-
changing customized kit training data between different DAWs
al sounding and has more crosstalk between the instruments
and systems or in order to build your own library of settings
so it may be more suitable to mixing situations. Gradually in-
matched to different recordings.
crease the setting to hard for more gated sounds which is use-
ful for extracting trigger signals from your drum mix (for ex-
Collapse/Expand GUI
ample).
Moving the mouse cursor to the upper left corner of the GUI
Smoothing brings up an icon that enables the analyzer section of to be
collapsed. This is a global preference saved for all instances.
The built in envelope smoothing feature can be used to con-
ceal separation artifacts which may be helpful when strongly
boosting or cutting individual drum instruments in your mix. PER CHANNEL CONTROLS
The effect of this feature can be visualized best in the ana-
lyzer’s envelope view. Active
The active button can be used to exclude a specific instrument
View Mode
from the separation. e.g. Disable the Tom and Floor tom chan-
Use the analyzer view mode setting to select between real- nels if you're processing a recording that does not contain any
time spectral analysis of your separated instruments and the tom hits. You can also use the active button to A/B the pro-
master signal or use the envelope view to monitor the signals' cessing currently applied to the instrument.
envelopes in the time domain. The reference spectra used for
isolating the individual instruments are shown as dark colored M and S buttons
lines, while the real-time spectrum is displayed in a solid color
Depress the M or S buttons to mute or solo individual sources
framed by a brighter line.
in your mix. By default the solo (S) buttons work cumulatively,
hold down the shift key for exclusive solo mode.

2
Kit Selection Menu ments if necessary. The learn buttons can be set to latch mode
by holding down the shift key while clicking. If you have acci-
Use the drop down menu below the active button to select
dentally hit one of the learn buttons you can undo that action
between four different preset kits suited to different genres,
by Ctrl/CMD + clicking the respective button.
recording styles and drum kits. It's recommended to solo (see
below) and evaluate the separation on each instrument to find
out which of the presets works best for your material. The cus- EQ / Dyn / Vol Buttons (Master Channel)
tom option corresponds to the setting trained using the learn Use the mixer mode buttons to toggle between the mixer's
mode (see below). Hold down the shift key while clicking the Volume, EQ and Dynamics panels.
selection menu to set the desired kit on all channels.
PER CHANNEL CONTROLS (VOLUME PAGE)
Learn
Since it's virtually impossible to offer presets that work for any Pan Fader
type of genre, playing style, recording setup, drum kit, set of The pan fader allows you to adjust the L/R balance of each in-
skins, cymbals etc., Drums SSX offers a learn mode to adapt dividual instrument as well as the master signal. This control is
each channel to your material as good as possible. Reaching only available in stereo instances of the plugin.
for the learn mode button is recommended whenever you're
unsatisfied with the separation result or whenever there are
Volume
any custom instruments in your recording such as woodblocks
or cowbells. e.g. To train a custom kick drum, try to find a spot Drag the volume fader to adjust the gain of each channel in a
in the played part that contains an isolated or at least very range from -60 thru +15 dB. Values can also be entered
dominant kick hit and setup a selection on that hit in your through the text field below the fader.
DAW. Then start your DAW's playback in loop mode and de-
press the learn button on the kick channel. Drums SSX will PP Meter
then average the spectrum of the incoming signal which will The meter right of the volume fader monitors the peak pro-
substantially improve the separation result for the trained in- gram level in dB FS of each respective channel.
strument. Repeat the described steps for all the other instru-

4
PER CHANNEL CONTROLS (EQ PAGE) PER CHANNEL CONTROLS (DYNAMICS PAGE)

H (High Shelf Gain) Att (Attack)

The H knob controls the amount of high shelf gain for each in- The attack knob allows you to emphasize or attenuate the at-
dividual channel between -15 thru +15 dB. The shelving fre- tack phase of each channel in the range between -15 thru +15
quency is fixed at 10 kHz. Values can also be entered using the dB.
text field below the knob.
Sus (Sustain)
M (Mid Band Gain) The sustain knob allows you to emphasize or attenuate the
The M knob controls the amount of mid band gain of each in- sustain phase of each instrument in the range between -15
dividual channel in the range from -15 thru +15 dB. The mid thru +15 dB.
band is a sweepable, proportional Q, peaking filter. Values can
also be entered using the text field below the knob. Sat (Saturation)
Use the saturation knob to add subtle THD to each individual
MF (Mid Band Frequency) channel. The the range is 0 thru 100%.
This knob sets the center frequency of each associated mid
band filter in the range from 220 thru 5.5 kHz. Values can also Mix
be entered using the text field below the knob.
The mix knob allows to dial in a blend between the dry version
of each individual channel signal and the version processed
L (Low Shelf Gain)
according to the adjustments you made on the dynamics
The L knob controls the amount of low shelf gain for each indi- page.
vidual channel between -15 thru +15 dB. The shelving fre-
quency is fixed at 100 Hz. Values can also be entered using the
text field below the knob.

6
UNMIXING APPLICATIONS track even if they were captured using multiple microphones
(e.g. recorder-man, XY overheads and a kick mic etc.).

Drums SSX can be run in a Stereo → 5.1 or 5.1 → 5.1 unmixing

Post production either took place on the recorder‘s built in
mode if those configurations are supported by your DAW. The
mixer or ITB on a friend’s computer, who was one of the few
incoming stereo L/R signals will be unmixed according to the
guys I knew that was already capable of transferring the tape
following hardwired routing:
tracks into the digital domain and I still have many of those re-
cordings floating around on various hard drives. Being a drum-
Cymbal channel (left) → 5.1 L speaker
mer myself I always wished for a little more granularity that
Cymbal channel (right) → 5.1 R speaker
goes beyond the usual tools like EQ, dynamic or even multi-
Snare channel (mono) → 5.1 C speaker
band processors given they had to be applied to only a single
Kick channel (mono) → 5.1 Lfe speaker
mono drum track.
Tom + Floortom channels (left) → 5.1 Ls speaker
Tom + Floortom channels (right) → 5.1 Rs speaker
Some years later, during my diploma thesis, I also happened
to do a fair amount of research on blind source separation
E.g. Pro Tools allows for bouncing the individual 5.1 channels
(BSS) which is the problem of separating a mix of two or more
to multiple wav tracks so they can be re-imported for arbitrary
audio sources into its components without any prior know-
post processing just as if you were working with a multitrack
ledge. This is how the idea grew to employ some of the ap-
recording.
proaches used in BSS in order to isolate the single instruments
from a mono or stereo drum signal as good as possible so
THE STORY BEHIND DRUMS SSX they can be processed individually using the same tool set that
would be applied to a multitrack drum recording.
My first introduction to multitrack recording systems was a
Portastudio that had to offer an 8 channel mixer and allowed
for recording up to 4 tracks simultaneously which were printed
to cassette tape. Whenever it came to recording say a typical
rock band setting, the drums had to be mixed down to a mono

8
A BRIEF ESSAY ON THE DRUM SSX' CORE ALGORITHM columns of B the NMF algorithm will iteratively “learn” a set of
B and G that minimize the distance between X and B•G , typic-
Drum SSX' separation method is based on an online variant ally according to the Euclidean distance or Kullback-Leibler di-
the non-negative matrix factorization (NMF) which simply put vergence. B0 corresponds to the preset kit spectra or
is a machine learning algorithm typically used in the separa- learned sound sources that model the average excitation of
tion of sound signals into their components. each individual instrument.

First, lets consider a standard windowed short time Fourier Each “layer” of B•G, that in the online context is simply a
transform (STFT) → inverse short time Fourier transform (IST - column of B weighted by the respective component in G,
FT) → overlap add signal chain. The sound signal mixture to be which track the activity envelopes over time, can be used to
separated is modeled as a set of spectra that vary over time. In obtain a time variant Wiener filter mask H. These masks can
the spectral domain this can be expressed as a linear be applied to the complex valued half spectrum S correspond-
combination of spectral basis vectors B that at each time in- ing to X in order to reconstruct estimates of the separated
stance are weighted according to an activation or envelope sound source spectra to be fed into the ISTFT:
vector G. Let X be the magnitude spectrum of the source sig-
nal at time instance k: Y ( k , m)=H (k , m)⋅S (m) , m∈{1.. M } .

X (k )≈ B(k )⋅G(k )
N ×1
X ∈ℝ ,
N ×M
B∈ℝ ,
G ∈ℝ M ×1

Where M corresponds to the number of modeled sources to

be separated and N is the spectral dimension of the spectral
frame to be analyzed.

Given a set of a priori spectra B0 initially given in the

10
WHY ARE THERE ONLY 5 SOURCES? ACKNOWLEDGEMENTS

In theory, it's possible to set any number of modeled sound Igor Korzhov of N.C. Rec
sources M. However, given that in practice there is always a
substantial amount of spectral overlap or crosstalk between for providing the training signals for the dry and metal kits.
e.g. the Hi-Hat and Overhead signals in typical recordings of
acoustic drums yielding unsatisfying separation results. As a Prof. Dr. Martin Spiertz
consequence of this the number of instruments has been cut
to the 5 most significant sound sources in order to optimize for supporting me during my thesis at IENT, Aachen, Germany and
the per se separation between the various instruments allow- for basically teaching me everything I know about Blind Audio
ing you to achieve reasonable results. Source Separation.

Thank You!

12
SYSTEM BLOCK DIAGRAM

13
KEYBOARD SHORTCUTS PLUGIN ACTIVATION

VST hosts You can test our plugins for 14 days without any restrictions.
Simply click the "Activate Trial" button in the plugin GUI after
Ctrl/CMD+Click: Reset to default
the installation to start your trial period.
Alt+Drag: Switch to circular knob mode
Shift+Drag/Mouse wheel: Fine control
To activate your full license you can purchase a product activa-
tion key on our website. The activation key will be sent to you
AU / AAX hosts
by email once the payment process has been completed, suc-
Alt+Click: Reset to default cessfully. Click "Activate plugin..." in the plugin GUI or the drop
Shift+Drag/Mouse wheel: Fine control down menu below, enter your email and product activation
key and hit the "Activate" button to proceed.
All hosts
Double Click: Reset to default / jump to value
FAQS AND TECHNICAL SUPPORT
SYSTEM REQUIREMENTS
An overview of Frequently Asked Questions can be found on
• Intel compatible or Apple Silicon CPU our website under the following link
• 2GB RAM
• Mac OS 10.9 or newer (Mac users) https://fuseaudiolabs.com/faq.html
• Windows 7 or newer (PC users)
• A display resolution of 1280x1024 pixels or more Our tech support can be reached via the following link
• A VST2, VST3, AAX or AU compatible 64-bit host (Mac
users) https://fuseaudiolabs.com/support.html
• A VST2, VST3 or AAX compatible 32-bit or 64-bit host (PC
users)

14