CLAS/NR - Compressive Loudness Audio Shaping with Noise Reduction
The CLAS plugin has been a huge smash with recording engineers and
artists alike. But one of the problems with this kind of processing,
when used at high relative treble enhancment levels as an exciter, is
an increase in high-frequency noise and hiss. This is particularly
troublesome for live sessions and old recordings.
To overcome this problem when recording, we have implemented a variant
of CLAS that incorporates a sliding-filter Noise Reduction unit ahead
of the CLAS processing. Analog implementations of sliding-filter noise
reduction have been used in the past to implement Dolby-B and Dolby-C
noise reduction encoding. We are doing it here, for the first time, in
digital form.
In practice, one simply adjusts the Threshold for noise reduction
until, under ambient noise levels, one sees the Noise Reduction meter
fully closed down. The two red-bars on the left extend all the way to
the bottom. At thresholds just below that point, you will see the
noise-reduction bars dancing up and down.
If Threshold is set too far below your actual noise levels, then of
course the noise-reduction bars retract all the way up, indicating that
it thinks signal is present. The same happens with any noise gate
technology.
Conversely, with this kind of noise reduction scheme, setting the
threshold too far above the actual noise level will produce "noise
modulation" on actual signals - a kind of stuttering sound. That
typically happens when your threshold is 20 dB or more above the actual
noise level, and it occurs because the sliding-filters are biting into
the top of your sound spectrum in an effort to eat that much "noise".
But when adjusted properly, and this is very easy to do with our Noise
Reduction bar graph, the sliding-filter noise reduction technique works
very well indeed. With Noise Reduction ahead of CLAS processing, you
can safely boost the highs without fear of undue noise degradation. As
soon as the signal falls away, the sliding-filter slams shut, keeping
noise at bay.
CLAS-NR uses upsampling ahead of the filtering for sample rates below
80 kHz, using a slightly different technique than our PLParEQ, but for
the same kind of quality improvements. Filtered results are downsampled
again to your system sample rate. The upsampling used is much lighter
weight than PLParEQ requiring much lower CPU loading. (Note: the free
CLAS plugin does not perform this high SR filtering.)
CLAS-NR uses 64-bit computation throughout, and it is 64-bit ready for
Sonar 5. For 24/32 bit systems, the 64-bit results are dithered back to
your host wordsize using our sophisticated Gaussian PDF dither.
CLAS-NR performs equally well in stereo or on mono tracks, using
autosensing of the host connections. (Some hosts do not advertise Mono
inserts properly).
CLAS-NR can accommodate all sample rates.
http://refinedaudiometrics.com
PLParEQ is a 10-band Phase-Linear Parametric Equalizer
of the highest quality. Each filter may
assume any of many different filter characteristics,
and operate in either traditional phase-warping mode,
or our phase-linear mode. It uses the same internal
DSP core as all of our other high-end products. Audio
streams are treated in either stereo or mono.
Individual filters can be applied to either both
stereo channels, middle only, or side only.
This PC VST plugin can operate in traditional 32-bit
mode (24-bit audio) or as a 64-bit plugin for
Cakewalks new Sonar 5. All internal processing is
carried out in double-precision 64-bit floating
point, regardless of external host mode.
For audio streams at sample rates below 80 kHz the
DSP engine internally upsamples with high-quality
Sinc interpolation, applies its filtering, and then
downsamples back to your system sample rate.
Available filter types include all of the
conventional resonant 2-pole lowpass, highpass,
bandpass, band-reject, all-pass, high and low
shelving, and the three Oxford variants of
peaking/dipping filters. Also included are 1-pole
lowpass and highpass, 4-pole resonant Butterworth
lowpass and highpass, 6-pole notch for killing 50/60
Hz and other troublesome spurs (very tight and
narrow), 1/F pinking filter, tunable 1/F lowpass (-3
dB/octave), tunable F-highpass (+3 dB/octave), and A,
B, and C weighting filters. The list of filters
continues to grow as we identify more useful
variations.
You can run all of the filters at native sample rates
higher than 80 kHz, and forego the internal
upsampling conversions. Upper limits on the sample
rate (> 96 kHz) are dicted primarily by your
computer's capacity, and your need for high quality
at the very lowest frequencies (below 100 Hz).
Individual filters can be operated as either
traditional phase-warping, or phase-linear. Our
algorithms employ blocked processing for phase
linear operation, and produce phase linearity by
sending the signal through each filter twice - once
in the forward time direction, and then again the the
time reversed direction, thereby unwinding the phase
back to zero.
The blocks are reassembled using very high-quality
windowing and 8-fold temporal overlap. IMD resulting
from phase-linear operation has been measured as -120
dB from peak signal levels. (IMD test measurements*)
Corner frequencies for filters can be adjusted from
10 Hz to 30 kHz at all sample rates.
Computation proceeds in double-precision floating
point throughout the entire DSP core, which uses 56
bit mantissas and 80-bit intermediate products and
sums. At the last stage of conversion back to 24-bit
audio, we dither with a carefully crafted white
Gaussian dither from the 56-bit mantissas to the 24
bit mantissas utilized thorughout by VST hosts. Our
noise floor is typically measured at around -180
dB/Root(Hz). The Gaussian dither distribution is
provably more preserving than more commonly used
triangular distributions.
Our tests indicate that PLParEQ requires about 5-6%
of our computer speed capacity, with each additional
filter enabled requiring an additional 0.5-1% at the
highest quality levels. These tests were performed on
a 3 GHz Pentium-4 computer with HyperThreading
enabled. Different VST hosts will show varying
requirements. Our tests were performed with Mackie's
Tracktion 2 as the VST host. Performance on a more
modern Pentium IV are likely to show improvements
upon these results.
*(IMD test measurement statistics - The signal was a
0.5 dBFS 1 kHz sinewave at 44.1 kHz SR. The
filtering was performed as a Q=20 phase -linear
bandpass filter centered at 1 kHz with unity gain,
and using the highest quality setting = largest
resampling blocksize.
The test chart exhibits a spectral resolution of 0.67
Hz, with the IMD spurs spaced by 43.1 Hz. The signal
was internally upsampled to 88.2 kHz for filter
application, then downsampled back to 44.1 kHz for
final output.)
http://refinedaudiometrics.com
PLParEQX - Our Phase-Linear Parametric Equalizer
Featuring X-Technology for Light CPU Loading
PLParEQX utilize our X-Technology to provide phase
linear filtering at a very light CPU loading for
Intel and AMD computers.
Whereas PLParEQ is an ultra-high quality phase-linear
parametric equalizer for mastering use, the X
Technology series offers slightly lower quality at
significantly reduced CPU loading, making them useful
for routine mixing and tracking applications.
PLParEQX is a 10-band version. Both offer high
quality oversampling in the processing and up to 5
levels of Quality. The lower quality levels are
useful when you need minimum processing latency, and
in turn you are willing to accept increased levels of
sonic artifacts. Our measurements indicate that the
lowest quality level has IMD products about 96 dB
below signal levels.
Our X-Series filters are 64-bit ready for Sonar 5,
and offer classical filtering shapes in purely phase
linear mode for minimum sound degradation,
especially in sharp attacks when imposing strongly
varying filtering with frequency.
We offer a full complement of lowpass, bandpass,
band-reject, and highpass filters, classical shelving
and peaking filters, as well as some novel low-slope
filtering for when you just need a slight attenuation
with frequency. Also offered are A, B, and C
weighting filters, and a killer 6-pole Notch filter
for absolutely killing that bothersome spur.
The 3-band unit offers the basic complement of
filters, without the more advanced ones offered in
the 10-band unit.
The magic of our X-Technology is such that no matter
how many filters you activate, the CPU loading is the
same low level. On tests with a 3 GHz Pentium 4 we
find the loading in Sonar 5, running in 64-bit mode,
to be around 4% -- significantly lower than the
premier PLParEQ filters. The quality may be a bit
compromised, but can you hear the difference?
http://refinedaudiometrics.com
Download File Size:5.27 MB
|
