© 2025 David Flater
GPLv3
Dynamic Range Destroyer (DRD.lv2 or simply DRD) is an audio compressor plugin that aims to remove changes in volume from an audio stream. This effect is also called auto volume levelling, volume normalization, or "stable volume." It supports mono, stereo, 6-channel, and 8-channel streams, and it preserves the balance among channels.
The Makefile takes two variables:
# Prefix under which the LV2 include files are found (not necessarily where
# plugin bundles are installed)
LV2PREFIX ?= /usr
# Directory into which the plugin bundle (DRD.lv2) should be installed
# For a user install say DESTDIR=$HOME/.lv2
DESTDIR ?= /usr/local/lib/lv2
Example installation to home directory:
$ DESTDIR=$HOME/.lv2 make install
gcc -I/usr/include -O2 -fPIC -fvisibility=hidden -shared -lm -o DRD.so DRD.c
install -d /home/dave/.lv2/DRD.lv2
install DRD.so /home/dave/.lv2/DRD.lv2
install -m 644 manifest.ttl DRD.ttl /home/dave/.lv2/DRD.lv2
$ lv2ls
http://flaterco.com/lv2/DRD#6ch
http://flaterco.com/lv2/DRD#8ch
http://flaterco.com/lv2/DRD#mono
http://flaterco.com/lv2/DRD#stereo
If the plugin is not found by lv2ls, you need to set or change the value of the environment variable LV2_PATH. The default search path is supposed to be $HOME/.lv2:/usr/local/lib/lv2:/usr/lib/lv2.
Range (0 to ∞) s
Default 0.01 s
An indication of the rate at which DRD responds to increases in the volume of the input. It is an exponential process. The attack is the time to respond to 1 − 1/e or approximately 63% of an increase of input level.
Note: A long attack time will make DRD slow to respond when the volume increases, potentially resulting in more and longer periods of excessive output volume and clipping.
Range (0 to ∞) s
Default 0.5 s
An indication of the rate at which DRD responds to decreases in the volume of the input. It is an exponential process. The decay is the time to respond to 1 − 1/e or approximately 63% of a decrease of input level.
DRD can be installed as a global system audio filter for Linux with PipeWire. See https://flaterco.com/kb/audio/pipewire/volume.html for details and an installation kit.
If FFmpeg was configured with --enable-lv2, you can
use DRD through a command line option similar to -af 'lv2=p=http\\://flaterco.com/lv2/DRD#stereo:c=attack=0.01|decay=0.5'. Be
sure to match the plugin URI to the number of channels of the input file
(mono, stereo, 6ch, or 8ch) to avoid surprising behavior.
If mpv is able to find the run-time dependency
libavfilter, you can use DRD through a command line option similar to
--af='lavfi="lv2=p=http\\://flaterco.com/lv2/DRD#stereo:c=attack=0.01|decay=0.5"'.
Again, please match the plugin URI to the number of channels of the input
file to avoid surprising behavior.
Ardour works with DRD. It provides sliders to change attack and decay on the fly, albeit with a defaulted maximum of 1 s each.
Tenacity does not work with DRD at the moment. Tenacity supports LV2 plugins and is seemingly able to apply DRD, but the result is silent audio. The pertinent code in src/effects/lv2 is complex.
DRD has a completely flat response in the frequency domain, but perceived loudness fades at high frequencies at a rate that varies from person to person. So when the spectrum of the input is dominated by high frequencies—for example, from a high-pitched rattle or the compression artifacts that it begets—the listener may perceive an apparent dropout.
A simple cure for these apparent dropouts is to pre-filter out the poorly heard high frequencies. DRD will then set the volume based on the sound that you actually hear and no dropout will occur.
Here are two low-pass filters that work well for this purpose:
- With FFmpeg/libavfilter, 10 ms delay, close to brick-wall:
firequalizer=gain=if(lt(f\,7500)\,0\,-INF) - With Linux Studio Plugins, no delay, very steep:
lv2=p=http\\://lsp-plug.in/plugins/lv2/filter_stereo:c=f=7500|ft=0|mode=0|fm=4|s=7
DRD is functionally similar to the compand filter of FFmpeg with fixed parameters points=-100/-100|-50/-15|0/-15 soft-knee=1 gain=0 volume=-15 delay=0. The major differences are (1) balance is preserved (the same gain is applied to every channel) and (2) the transfer function is different (see plots below). Although these plots stop at 0 dB as the nominal maximum input volume, DRD effectively extends the straight lines above −50 dB as far as necessary to handle out-of-range input.
The process involves a state variable herein called the floating volume level. It follows the volume of the input in a general sense, effectively finding the average over a short time interval.
The following steps are done for each sample:
- Estimate the overall volume level of the input as the maximum level of any channel (absolute value).
- Update the floating volume level using a fraction of its difference from the volume level of the input. The fraction is derived from the attack or decay parameter (as applicable) as 1 − e^(−1/(parameter × sample rate)).
- Evaluate the gain function using the floating volume level as input.
- Apply that gain to every channel.
DRD does not avoid or prevent clipping, nor does it clip the output values itself. They are delivered as floats, which can go outside the intended range of −1 to 1. It is assumed and intended that clipping will occur downstream.
The plot below shows the nominal worst-case scenario where the input level instantaneously transitions from 0 (minimum) to 1 (maximum) at time 0 and stays there. Since the floating volume level sweeps the range starting from 0, it passes through the peak of maximum gain on its way to the top, resulting in an extreme pop 38.3 dB beyond the limit within the first few samples. Negative gain is applied after ⅕ of the attack time has elapsed.
Of course, if the input to DRD is already out of range, the excursion can be worse yet, but DRD will continue to reduce the gain until the −15 dB target is reached.
The starting point for DRD was eg-amp.lv2 in lv2 version 1.18.10. Comments from eg-amp remain in the source.
A simple example of a basic LV2 plugin with no additional features.
Copyright 2006–2016 David Robillard
Copyright 2006 Steve Harris
ISC license
The attack/decay and update_volume code was copied from af_compand.c of FFmpeg version 7.1.1 (libavfilter.so.10.4.100).
Copyright (c) 1999 Chris Bagwell
Copyright (c) 1999 Nick Bailey
Copyright (c) 2007 Rob Sykes
Copyright (c) 2013 Paul B Mahol
Copyright (c) 2014 Andrew Kelley
LGPL v2.1
Additional LV2 infrastructure followed the example of dpl.lv2 version 0.7.0.
Digital Peak Limiter
Copyright (C) 2018 Robin Gareus
GPL v2 and GPLv3
The release numbers X.Y used for github releases indicate the LV2 minorVersion and microVersion of the plugin. The LV2 versioning scheme is described as follows (summarizing from minorVersion):
The minor version MUST be incremented when backwards (but not forwards) compatible additions are made, for example the addition of a port to a plugin.
The micro version is incremented for changes which do not affect compatibility at all, for example bug fixes or documentation updates.
There is deliberately no major version: all versions with the same URI are compatible by definition. Replacing a resource with a newer version of that resource MUST NOT break anything. If a change violates this rule, then the URI of the resource (which serves as the major version) MUST be changed.
An odd minor or micro version, or minor version zero, indicates that the resource is a development version.
The minor and micro version are specified in the Turtle file DRD.ttl that is installed as part of the DRD.lv2 bundle.
The shared library is always installed as simply DRD.so. The version of
DRD.so can be retrieved from the binary with
strings DRD.so | grep -F 'DRD version'.