Your company here, and a link to your site. Click to find out more.

dsp - Man Page

an audio processing program with an interactive mode


dsp [options] path ... [!] [:channel_selector] [@[~/]effects_file] [effect [args ...]] ...


dsp is an audio processing program with an interactive mode.


Global options


Show help text.

-b frames

Set buffer size (must be given before the first input).

-R ratio

Set codec maximum buffer ratio (must be given before the first input).


Force interactive mode.


Disable interactive mode.


Disable progress display.


Silent mode.


Verbose mode.


Force dithering.


Disable dithering.


Don't drain effects chain before rebuilding.


Plot effects chain instead of processing audio.


Enable verbose progress display.


Use `sequence' input combining mode.

Input/output options



-t type


-e encoding



Big/little/native endian.

-r frequency[k]

Sample rate.

-c channels

Number of channels.


Equivalent to

	-t null null.

Inputs and Outputs

For a complete list of supported input/output types, run

	$ dsp -h

Input combining modes

In concatenate mode (the default), the inputs are concatenated in the order given and sent to the output. All inputs must have the same sample rate and number of channels.

In sequence mode, the inputs are sent serially to the output like concatenate mode, but the inputs do not need to have the same sample rate or number of channels. The effects chain and/or output will be rebuilt/reopened when required. Note that if the output is a file, the file will be truncated if it is reopened. This mode is most useful when the output is an audio device, but can also be used to concatenate inputs with different sample rates and/or numbers of channels into a single output file when used with the resample and/or remix effects.

Signal generator

The sgen input type is a basic (for now, at least) signal generator that can generate impulses and exponential sine sweeps. The syntax for the path argument is as follows:


type may be `sine' for sine sweeps or tones, or `delta' for a delta function (impulse). `sine' accepts the following arguments:


Frequency. If len is set and f1 is given, an exponential sine sweep is generated.

The arguments for `delta' are:


Offset in seconds, miliseconds or samples.


	$ dsp -t sgen -c 2 sine@0:freq=500-1k/sine@1:freq=300-800+2 gain -10


Full effects list

lowpass_1 f0[k]

Single-pole lowpass filter.

highpass_1 f0[k]

Single-pole highpass filter.

lowpass f0[k] width[q|o|h|k]

Double-pole lowpass filter.

highpass f0[k] width[q|o|h|k]

Double-pole highpass filter.

bandpass_skirt f0[k] width[q|o|h|k]

Double-pole bandpass filter with constant skirt gain.

bandpass_peak f0[k] width[q|o|h|k]

Double-pole bandpass filter with constant peak gain.

notch f0[k] width[q|o|h|k]

Double-pole notch filter.

allpass f0[k] width[q|o|h|k]

Double-pole allpass filter.

eq f0[k] width[q|o|h|k] gain

Double-pole peaking filter.

lowshelf f0[k] width[q|s|d|o|h|k] gain

Double-pole lowshelf filter.

highshelf f0[k] width[q|s|d|o|h|k] gain

Double-pole highshelf filter.

linkwitz_transform fz[k] qz fp[k] qp

Linkwitz transform (see http://www.linkwitzlab.com/filters.htm#9).


Compact Disc de-emphasis filter.

biquad b0 b1 b2 a0 a1 a2

Biquad filter.

gain [channel] gain

Gain adjustment. Ignores the channel selector when the channel argument is given.

mult [channel] multiplier

Multiplies each sample by multiplier. Ignores the channel selector when the channel argument is given.

add [channel] value

Applies a DC shift. Ignores the channel selector when the channel argument is given.

crossfeed f0[k] separation

Simple crossfeed for headphones. Very similar to Linkwitz/Meier/CMoy/bs2b crossfeed.

matrix4 [[options] surround_level]

2-to-4 channel (2 front and 2 surround) active matrix upmixer designed for plain (i.e. unencoded) stereo material. The matrix coefficients and the main ideas behind the steering behavior come from David Griesinger's published works on matrix surround.

The intended speaker configuration is fronts at ±30° and surrounds between ±60° and ±120°. The surround speakers must be calibrated correctly in level and frequency response for best results. No frequency contouring or delay is done internally, so it is highly recommended to apply delay and lowpass_1 effects to the surround outputs:

	matrix4 -6 :2,3 delay 15m lowpass_1 6k :

The settings shown above (-6dB surround level, 15ms delay, and 6kHz rolloff) are a good starting point, but may be adjusted to taste. The default surround_level is -6dB.

The front outputs replace the original input channels and the surround outputs are appended to the end of the channel list.

Options are given as a comma-separated list. Recognized options are:


Disable directional boost of front channels.


Show a status line (slightly broken currently, but still useful for debugging).


Toggle the effect when `effect.signal()' is called.

remix channel_selector|. ...

Select and mix input channels into output channels. Each channel selector specifies the input channels to be mixed to produce each output channel. `.' selects no input channels. For example,

	remix 0,1 2,3

mixes input channels 0 and 1 into output channel 0, and input channels 2 and 3 into output channel 1.

	remix -

mixes all input channels into a single output channel.


Convert stereo to mid/side.


Convert mid/side to stereo.

delay delay[s|m|S]

Delay line. The unit for the delay argument depends on the suffix used: `s' is seconds (the default), `m' is milliseconds, and `S' is samples.

resample [bandwidth] fs[k]

Sinc resampler. Ignores the channel selector.

fir [~/]filter_path

Non-partitioned 64-bit FFT convolution. Latency is equal to the length of the filter.

fir_p [min_part_len [max_part_len]] [~/]filter_path

Non-uniform partitioned 64-bit FFT convolution. Runs slower than the zita_convolver effect, but potentially useful if you need more precision and/or lower latency. Latency is equal to min_part_len (16 samples by default). {min,max}_part_len must be powers of 2.

hilbert taps

Simple FIR approximation of a Hilbert transform. The number of taps must be odd. Bandwidth is controlled by the number of taps.

decorrelate [stages]

Allpass decorrelator as described in “Frequency-Dependent Schroeder Allpass Filters” by Sebastian J. Schlecht (doi:10.3390/app10010187). Default number of stages is 5. The average group delay with the default settings is about 8.5ms at high frequencies.

zita_convolver [min_part_len [max_part_len]] [~/]filter_path

Partitioned 32-bit FFT convolution using the zita-convolver library. Latency is equal to min_part_len (64 samples by default). {min,max}_part_len must be powers of 2 between 64 and 8192.

noise level

Add TPDF noise. The level argument specifies the peak level of the noise (dBFS).

ladspa_host module_path plugin_label [control ...]

Apply a LADSPA plugin. Supports any number of input/output ports (with the exception of zero output ports). Plugins with zero input ports will replace selected input channels with their output(s). If a plugin has one or zero input ports, it will be instantiated multiple times to handle multi-channel input.

Controls which are not explicitly set or are set to `-' will use default values (if available).

The `LADSPA_PATH' environment variable can be used to set the search path for plugins.

stats [ref_level]

Display the DC offset, minimum, maximum, peak level (dBFS), RMS level (dBFS), crest factor (dB), peak count, peak sample, number of samples, and length (s) for each channel. If ref_level is given, peak and RMS levels relative to ref_level will be shown as well (dBr).

Exclamation mark

A `!' marks the effect that follows as `non-essential'. If an effect is marked non-essential and it fails to initialize, it will be skipped.

Selector syntax


2-2 to n
-40 through 4
1,31 and 3
1-4,7,9-1 through 4, 7, and 9 to n

Width suffixes

qQ-factor (default).
sSlope (shelving filters only).
dSlope in dB/octave (shelving filters only).
oBandwidth in octaves.
hBandwidth in Hz.
kBandwidth in kHz.

Note: The `d' width suffix also changes the definition of f0 from center frequency to corner frequency (like Room EQ Wizard and the Behringer DCX2496).

File paths

  • On the command line, relative paths are relative to `$PWD'.
  • Within an effects file, relative paths are relative to the directory containing said effects file.
  • The `~/' prefix will be expanded to the contents of `$HOME'.

Effects file syntax

  • Arguments are delimited by whitespace.
  • If the first non-whitespace character in a line is `#', the line is ignored.
  • The `\' character removes any special meaning of the next character.


	gain -10
	# This is a comment
	eq 1k 1.0 +10.0 eq 3k 3.0 -4.0
	lowshelf 90 0.7 +4.0

Effects files inherit a copy of the current channel selector. In other words, if an effects chain is this:

	:2,4 @eq_file.txt eq 2k 1.0 -2.0

eq_file.txt will inherit the `2,4' selector, but any selector specified within eq_file.txt will not affect the `eq 2k 1.0 -2.0' effect that comes after it.


Read file.flac, apply a bass boost, and write to alsa device hw:2:

	dsp file.flac -ot alsa -e s24_3 hw:2 lowshelf 60 0.5 +4.0

Plot amplitude vs frequency for a complex effects chain:

	dsp -pn gain -1.5 lowshelf 60 0.7 +7.8 eq 50 2.0 -2.7 eq 100 2.0 -3.9
	  eq 242 1.0 -3.8 eq 628 2.0 +2.1 eq 700 1.5 -1.0
	  lowshelf 1420 0.68 -12.5 eq 2500 1.3 +3.0 eq 3000 8.0 -1.8
	  eq 3500 2.5 +1.4 eq 6000 1.1 -3.4 eq 9000 1.8 -5.6
	  highshelf 10000 0.7 -0.5 | gnuplot

Implement an LR4 crossover at 2.2KHz, where output channels 0 and 2 are the left and right woofers, and channels 1 and 3 are the left and right tweeters, respectively:

	dsp stereo_file.flac -ot alsa -e s32 hw:3 remix 0 0 1 1 :0,2
	  lowpass 2.2k 0.707 lowpass 2.2k 0.707 :1,3 highpass 2.2k 0.707
	  highpass 2.2k 0.707 :

Apply effects from a file:

	dsp file.flac @eq.txt

Ladspa Frontend


ladspa_dsp looks for configuration files in the following directories:

  • $XDG_CONFIG_HOME/ladspa_dsp
  • $HOME/.config/ladspa_dsp (if $XDG_CONFIG_HOME is not set)
  • /etc/ladspa_dsp

To override the default directories, set the `LADSPA_DSP_CONFIG_PATH' environment variable to the desired path(s) (colon-separated).

Each file that is named either config or config_<name> (where <name> is any string) is loaded as a separate plugin. The plugin label is either ladspa_dsp (for config) or ladspa_dsp:<name> (for config_<name>).

Configuration files are a simple key-value format. Leading whitespace is ignored. The valid keys are:


Number of input channels. Default value is 1. May be left unset unless you want individual control over each channel.


Number of output channels. Default value is 1. Initialization will fail if this value is set incorrectly.


Set `LC_NUMERIC' to the given value while building the effects chain. If the decimal separator defined by your system locale is something other than `.', you should set this to `C' (to use `.' as the decimal separator) or an empty value (to use the decimal separator defined by your locale).


String to build the effects chain. The format is the same as an effects file, but only a single line is interpreted.

Example configuration:

	# This is a comment
	effects_chain=gain -3.0 lowshelf 100 1.0s +3.0 @/path/to/eq_file

Relative file paths in the effects_chain line are relative to the directory in which the configuration file resides.

The loglevel can be set to `VERBOSE', `NORMAL', or `SILENT' through the `LADSPA_DSP_LOGLEVEL' environment variable.

Note: The resample effect cannot be used with the LADSPA frontend.


See https://github.com/bmc0/dsp/blob/master/README.md for usage examples.


No support for metadata.

Some effects do not support plotting.


This software is released under the ISC license.