About

dsp is an audio processing program with an interactive mode.

Building

Dependencies

• GNU Make
• pkg-config

Optional dependencies

• fftw3: For resample, fir, fir_p, and hilbert effects.
• zita-convolver: For the zita_convolver effect.
• libsndfile: For sndfile input/output support (recommended).
• ffmpeg (libavcodec, libavformat, and libavutil): For ffmpeg input support.
• alsa-lib: For alsa input/output support.
• libao: For ao output support.
• libmad: For mp3 input support.
• libpulse-simple: For PulseAudio input/ouput support.
• LADSPA: For the LADSPA frontend and the ladspa_host effect.
• libltdl (libtool): For the ladspa_host effect.

Build

$ make

Run ./configure [options] manually if you want to build with non-default options. Run ./configure --help to see all available options.

Install

# make install

Synopsis

dsp [options] path ... [effect [args]] ...

Options

Global options

Flag	Description
-h	Show help text.
-b frames	Block size (must be given before the first input).
-i	Force interactive mode.
-I	Disable interactive mode.
-q	Disable progress display.
-s	Silent mode.
-v	Verbose mode.
-d	Force dithering.
-D	Disable dithering.
-E	Don't drain effects chain before rebuilding.
-p	Plot effects chain magnitude response instead of processing audio.
-P	Same as -p, but also plot phase response.
-V	Verbose progress display.
-S	Use "sequence" input combining mode.

Input/output options

Flag	Description
-o	Output.
-t type	Type.
-e encoding	Encoding.
-B/L/N	Big/little/native endian.
-r frequency[k]	Sample rate.
-c channels	Number of channels.
-R ratio	Buffer ratio.
-n	Equivalent to -t null null.

Inputs and Outputs

Supported input/output types

Type	Modes	Encodings
null	rw	sample_t
sgen	r	sample_t
sndfile	r	autodetected
wav	rw	s16 u8 s24 s32 float double mu-law a-law ima_adpcm ms_adpcm gsm6.10 g721_32
aiff	rw	s16 s8 u8 s24 s32 float double mu-law a-law ima_adpcm gsm6.10 dwvw_12 dwvw_16 dwvw_24
au	rw	s16 s8 s24 s32 float double mu-law a-law g721_32 g723_24 g723_40
raw	rw	s16 s8 u8 s24 s32 float double mu-law a-law gsm6.10 vox_adpcm dwvw_12 dwvw_16 dwvw_24
paf	rw	s16 s8 s24
svx	rw	s16 s8
nist	rw	s16 s8 s24 s32 mu-law a-law
voc	rw	s16 u8 mu-law a-law
ircam	rw	s16 s32 float mu-law a-law
w64	rw	s16 u8 s24 s32 float double mu-law a-law ima_adpcm ms_adpcm gsm6.10
mat4	rw	s16 s32 float double
mat5	rw	s16 u8 s32 float double
pvf	rw	s16 s8 s32
xi	rw	dpcm_8 dpcm_16
htk	rw	s16
sds	rw	s16 s8 s24
avr	rw	s16 s8 u8
wavex	rw	s16 u8 s24 s32 float double mu-law a-law
sd2	rw	s16 s8 s24
flac	rw	s16 s8 s24
caf	rw	s16 s8 s24 s32 float double mu-law a-law
wve	rw	a-law
ogg	rw	vorbis
mpc2k	rw	s16
rf64	rw	s16 u8 s24 s32 float double mu-law a-law
ffmpeg	r	autodetected
alsa	rw	s16 u8 s8 s24 s24_3 s32 float double
ao	w	s16 u8 s32
mp3	r	mad_f
pcm	rw	s16 u8 s8 s24 s32 float double
pulse	rw	s16 u8 s24 s24_3 s32 float

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[@channel_selector][:arg[=value]...]][/type...][+len[s|m|S]]

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

• freq=f0[k][-f1[k]] Frequency. If len is set and f1 is given, an exponential sine sweep is generated.

The arguments for delta are:

• offset=time[s|m|S] Offset in seconds, miliseconds or samples.

Example:

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

Effects

Complete effects list

• lowpass_1 f0[k]
  First-order lowpass filter.

• highpass_1 f0[k]
  First-order highpass filter.

• allpass_1 f0[k]
  First-order allpass filter.

• lowshelf_1 f0[k] gain
  First-order lowshelf filter.

• highshelf_1 f0[k] gain
  First-order highshelf filter.

• lowpass_1p f0[k]
  Single pole lowpass (EWMA) filter.

• lowpass f0[k] width[q|o|h|k]
  Second-order lowpass filter.

• highpass f0[k] width[q|o|h|k]
  Second-order highpass filter.

• bandpass_skirt f0[k] width[q|o|h|k]
  Second-order bandpass filter with constant skirt gain.

• bandpass_peak f0[k] width[q|o|h|k]
  Second-order bandpass filter with constant peak gain.

• notch f0[k] width[q|o|h|k]
  Second-order notch filter.

• allpass f0[k] width[q|o|h|k]
  Second-order allpass filter.

• eq f0[k] width[q|o|h|k] gain
  Second-order peaking filter.

• lowshelf f0[k] width[q|s|d|o|h|k] gain
  Second-order lowshelf filter.

• highshelf f0[k] width[q|s|d|o|h|k] gain
  Second-order highshelf filter.

• lowpass_transform fz[k] qz fp[k] qp
  Second-order lowpass transformation filter. Cancels the poles defined by fz and qz and replaces them with new poles defined by fp and qp. Gain is unity at DC.

• highpass_transform fz[k] qz fp[k] qp
  Second-order highpass transformation filter. Also known as a Linkwitz transform (see http://www.linkwitzlab.com/filters.htm#9). Same as lowpass_transform except the gain is unity at Fs/2.

• linkwitz_transform fz[k] qz fp[k] qp
  Alias for highpass_transform.

• deemph
  Compact Disc de-emphasis filter.

• biquad b0 b1 b2 a0 a1 a2
  Biquad filter.

• gain gain_dB
  Gain adjustment in decibels.

• mult multiplier
  Multiplies each sample by multiplier.

• add value
  Applies a DC shift.

• 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 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. The surrounds should be delayed by about 10-25ms (acoustically) relative to the fronts. No frequency contouring is done internally, so applying low pass and/or shelving filters to the surround outputs is recommended:

   matrix4 surround_delay=15m -6 :2,3 lowpass_1 10k :
  

  The settings shown above (-6dB surround level, 15ms delay, and 10kHz rolloff) are a good starting point, but may be adjusted to taste. The default surround_level is -6dB. Applying the decorrelate effect to the surround outputs (optionally with the -m flag) seems to further improve the spatial impression (note: adjust surround_delay to compensate for the decorrelate effect's group delay).

  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:

  • no_dir_boost
    Disable directional boost of front channels.
  • show_status
    Show a status line (slightly broken currently, but still useful for debugging).
  • signal
    Toggle the effect when effect.signal() is called.
  • linear_phase (matrix4_mb only)
    Apply an FIR filter to correct the phase distortion caused by the IIR filter bank. Has no effect with matrix4. Requires the fir effect.
  • surround_delay=delay[s|m|S]
    Surround output delay. Default is zero.

• matrix4_mb [options] [surround_level]
  Like the matrix4 effect, but divides the input into ten individually steered bands in order to improve separation of concurrent sound sources. See the matrix4 effect description for more information.

• remix selector|. ...
  Select and mix input channels into output channels. Each selector argument specifies the input channels to be mixed to produce an 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. The active channel selector is used as an input channel mask for the selector arguments.

• st2ms Convert stereo to mid/side.

• ms2st 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. If delay is negative, a positive delay is applied to all channels which are not selected (except when plotting—an actual negative delay is possible in that case).

• resample [bandwidth] fs[k]
  Sinc resampler. Ignores the channel selector.

• fir [file:][~/]filter_path|coefs:list[/list...]
  Non-partitioned 64-bit direct or FFT convolution. Latency is zero for filters up to 16 taps. For longer filters, the latency is equal to the fft_len reported in verbose mode. Each list is a comma-separated list of coefficients for one filter channel. Missing values are filled with zeros.

• fir_p [max_part_len] [file:][~/]filter_path|coefs:list[/list...]
  Zero-latency non-uniform partitioned 64-bit direct/FFT convolution. Runs slower than the zita_convolver effect, but useful if you need higher precision and/or zero latency. max_part_len must be a power of 2. Each list is a comma-separated list of coefficients for one filter channel. Missing values are filled with zeros.

• 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.

• hilbert [-p] taps
  Simple FIR approximation of a Hilbert transform. The number of taps must be odd. Bandwidth is controlled by the number of taps. If -p is given, the fir_p convolution engine is used instead of the default fir engine.

• decorrelate [-m] [stages]
  Allpass decorrelator as described in "Frequency-Dependent Schroeder Allpass Filters" by Sebastian J. Schlecht (doi:10.3390/app10010187). If -m is given, the same filter parameters are used for all input channels. The default number of stages is 5, which results in an average group delay of about 9.5ms at high frequencies.

• 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). 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).

Selector syntax

[[start][-[end]][,...]]

Example	Description
<empty>	all
-	all
2-	2 to n
-4	0 through 4
1,3	1 and 3
1-4,7,9-	1 through 4, 7, and 9 to n

Note: There is no difference between 1,3 and 3,1. Order is not preserved.

Filter width suffixes

Suffix	Description
q	Q-factor (default).
s	Slope (shelving filters only).
d	Slope in dB/octave (shelving filters only).
o	Bandwidth in octaves.
h	Bandwidth in Hz.
k	Bandwidth 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.

Channel selectors and masks

A colon (:) followed by a selector (see "Selector syntax") specifies the input channels for effects that follow. For example,

:0,2 eq 1k 1.0 -6

will apply an eq effect to channels 0 and 2. If an effect changes the total number of channels, the last channel selector given is parsed again. Additional channels are not added unless the selector includes an unbounded range.

Channel numbers refer to the channels in the active channel mask, which is a property of the containing block. Blocks may be created using braces ({ ... }) or by sourcing a file (see "Effects files"). The channel mask is derived from the active channel selector at creation. For example,

:1,3 { :0 gain -6 :1 gain +6 }

creates a block with the mask 1,3. Within the block, :0 selects the first channel in the mask (channel 1), and :1 selects the second channel in the mask (channel 3). Channel selectors have block scope.

Channels are automatically added or removed from the active channel mask if an effect changes the total number of channels. Additional channels are always appended to the end of the channel list.

Effects files

Files may be sourced using the @ directive: @[~/]path/to/file. See "File paths" for more information about how paths are interpreted. Note that sourcing a file implicitly creates a block (see "Channel selectors and masks"). Within a file, lines in which the first non-whitespace character is # are ignored. A backslash (\) may be used to escape whitespace, #, or \. Example:

gain -4.0
# This is a comment
lowshelf 90 1s +4 eq 3k 1.5 -3

Other directives

An exclamation mark (!) allows initialization failure of the effect that follows.

Examples

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

Plot the magnitude vs frequency response of an effects chain:

dsp -pn [effect [args]] ... | gnuplot

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

dsp stereo_file.flac -ot alsa -e s32 hw:3 remix 0 1 0 1
  :0,1 highpass 2.2k 0.7071 highpass 2.2k 0.7071 :
  :2,3 lowpass 2.2k 0.7071 lowpass 2.2k 0.7071 :

Apply effects from a file:

dsp file.flac @eq.txt

LADSPA frontend

Configuration

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:

• input_channels
  Number of input channels. Default value is 1. May be left unset unless you want individual control over each channel.
• output_channels
  Number of output channels. Default value is 1. Initialization will fail if this value is set incorrectly.
• LC_NUMERIC
  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).
• effects_chain
  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
input_channels=1
output_channels=1
LC_NUMERIC=C
effects_chain=gain -3 lowshelf 100 1s +3 @/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.

Usage example: Route alsa audio through ladspa_dsp

Put this in ~/.asoundrc:

pcm.dsp {
	type plug
	slave {
		format FLOAT
		rate unchanged
		channels unchanged
		pcm {
			type ladspa
			path "/usr/lib/ladspa"
			playback_plugins [{
				label "ladspa_dsp"
			}]
			slave.pcm {
				type plug
				slave {
					pcm "<hw_device>"
					rate unchanged
					channels unchanged
				}
			}
		}
	}
}

Replace <hw_device> with the preferred output device (hw:0, for example).

If you need individual control over each channel, you need to set the number of (output) channels:

pcm.dsp {
	type plug
	slave {
		format FLOAT
		rate unchanged
		pcm {
			type ladspa
			channels <channels>
			path "/usr/lib/ladspa"
			playback_plugins [{
				label "ladspa_dsp"
			}]
			slave.pcm {
				type plug
				slave {
					pcm "<hw_device>"
					rate unchanged
					channels unchanged
				}
			}
		}
	}
}

To make dsp the default device, append this to ~/.asoundrc:

pcm.!default {
	type copy
	slave.pcm "dsp"
}

Usage example: Route pulseaudio audio through ladspa_dsp (tested with Ubuntu 18.04; contributed by shaffenmeister)

1. Prepare .asoundrc as stated above.
2. Determine pulseaudio master sink using pacmd list sinks. Use attribute name of the pulseaudio sink you plan to use (e.g. alsa_output.pci-0000_00_14.2.analog-stereo).
3. Execute analyseplugin <path to LADSPA plugin>/ladspa_dsp.so to determine plugin name and label.
4. Run pacmd load-module module-ladspa-sink sink_name=ladspa_out sink_master=<master_sink> plugin=<plugin name> label=<plugin label>.
5. Select new LADSPA sink as system sink (Ubuntu 18.04 Desktop: Settings > Sound > Output > LADSPA_Plugin <plugin label> on <master sink>).

Example:

pacmd list sinks
analyseplugin /usr/local/lib/ladspa/ladspa_dsp.so
pacmd load-module module-ladspa-sink sink_name=ladspa_out sink_master=alsa_output.pci-0000_00_14.2.analog-stereo plugin=ladspa_dsp label=ladspa_dsp

Load LADSPA plugin as system default

To load the LADSPA module at system startup for all users include settings in /etc/pulse/default.pa:

.ifexists module-ladspa-sink.so
.nofail
load-module module-ladspa-sink sink_name=ladspa_out sink_master=<master_sink> plugin=<plugin name> label=<plugin label>
.fail
.endif

Load LADSPA plugin as user default

To load the LADSPA module at user login include settings in ~/.config/pulse/default.pa:

#!/usr/bin/pulseaudio -nF
.include /etc/pulse/default.pa
.ifexists module-ladspa-sink.so
.nofail
load-module module-ladspa-sink sink_name=ladspa_out sink_master=<master_sink> plugin=<plugin name> label=<plugin label>
.fail
.endif

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

Bugs

• No support for metadata.
• Some effects do not support plotting.
• When plotting an effects chain containing the noise effect, a different random sequence is generated for each output channel regardless of whether the noise should be correlated between outputs. Summing correlated noise works correctly.

License

This software is released under the ISC license.