Skip to main content
Topic: release of the Postfish (Read 29334 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

release of the Postfish

Not a really big deal, mainly because the Postfish is of admittedly limited appeal (it's targetted at creators, not consumers ;-)

...but to get it out of my head for a bit, I declare today's last Postfish commit to consitute a prerelease of this fine app.  I have to put it down for a while but it seems silly to keep something very useful hidden for no reason.

Have no freaking clue what the Postfish is?  Well, here's the prerelease README:

This README file covers the 2004-05-29 pre-release of the Postfish by
Xiph.Org.

>>>> What is the Postfish?

  The Postfish is a digital audio post-processing, restoration,
  filtering and mixdown tool.  It works as a linear audio filter, much
  like a rack of analog effects.  The first stage of the filter
  pipeline provides a bank of configurable per-channel processing
  filters for up to 32 input channels.  The second stage provides
  mixdown of the processed input audio into a group of up to eight
  output channels.  The third stage applies processing filters to the
  output group post-mixdown.

  The Postfish is a stream filter; feed it audio from a list of files
  or input stream, and it renders audio to standard out, as well as
  optionally providing a configurable audio playback monitor via a
  sound device.  If the input audio is being taken from files,
  Postfish also provides simple forward/back/cue seeking and A-B
  looping control.

>>>> What is the Postfish for?

  The Postfish intends to include exactly and only the most useful
  basic filters needed to produce a good mixdown from audio recorded
  'in the field'.  The filter set also comprises the fundamentals
  needed for master mixdown in a small studio.  It is not an editor;
  for that reason, it's intended to be used with an audio editor, such
  as Audacity.

  If, for example, you've just multi-track recorded a rehearsal of
  your troupe's current rock opera and the Director then appears out
  of nowhere (they always do) and says "Have a mix for my review by
  morning", the Postfish is all you need.

  Or, as another example, you've recorded for a band who'd like to put
  out a CD of the live performance...  All the band FX are already in
  the multi-track, so the Postfish plus Audacity is all you need.

  In a studio situation, tracks usually get recorded dry, so there's
  generally multiple mixdown stages of adding effects.  Postfish
  (obviously) does not give you a large array of instrument or
  situation-specific effects and it never will (dammit).  What it does
  give you is the effects necessary to take the tracks from earlier
  mixing and produce intermediate mixes and final masters.  Of course,
  if you already have $100k of analog rack... you likely won't be
  using Postfish.  But hey, who knows....

>>>> What effects does the Postfish include?

Declipper:

  The Postfish declipper is a 'build audio from scratch'
  reconstruction filter. Any section of audio exceeding a configured
  amplitude threshold is marked 'lost' and the filter builds new
  audio to fill the gap. In this way it can be used to repair both
  digital clipping that occurred during sampling, as well as analog
  clipping that may have happened at an earlier stage.

Single-band compander:

  A single-band compander is used to compress, limit, expand, or gate
  an input signal, thus providing basic dynamic range manipulation
  abilities. 

  The Postfish single-band compander provides independent over and
  under threshold controls for each channel, each providing expansion,
  compression, attack, release, lookahead and soft-knee configuration
  for three independent over, mid and under tracking filters.  Each
  filter may also be configured to track by peak, or track by RMS
  energy.

Multi band compander:

  The Postfish multi-band compander is similar to the single-band
  compander above, and provides all of the same controls with an
  addition: each over/under threshold is configurable by full-octave,
  half-octave or third-octave bands, for up to 30 bands of independent
  companding for each of 32 input channels and the group of eight
  outputs.

  The multi-band compander includes a global 'mid' compand slider,
  like the single-band compander.  This slider acquires a new use in
  multi-band mode however, where it can flatten or expand the dynamic
  range of an entire channel (or the entire recording) without
  any artifacts.

Equalization:

  30 band -60/+30dB 1/3-octave beat-less EQ per input channel and full
  output group.

Deverberator:

  Live recordings have a tendency to end up with too much reverb,
  especially when one is forced to use ambient miking.  The
  deverberator dries out overly 'wet' live signals.

Reverberator:

  ...for adding reverb to signals that are too 'dry', especially to
  even out apparent depth when mixing close miked signals (like
  vocals) to an ambient-miked signal.

  The Postfish provides a stereo reverb per input channel and a mono
  reverb per output post-mix.

Limiting:

  Simple, old-fashioned causal output hard-limiter to avoid unexpected
  digital clipping on the output.  Configurable by threshold, knee
  depth and release speed.

Mixdown:

  Postfish provides both a master attenuation and delay panel (which
  places these basic sliders for all channels on one window) as well
  as per-input mixdown that allows each input channel to be multiply
  routed to one or all of eight output through an additional cascade
  of four additional independent attenuation/delay/invert units per
  input, as well also allowing each input to be mixed through a
  'crossplacer'.

  The crossplacer is used to place any input into a stereo [or
  greater] image by altering not only the relative attenuation across
  channels (the 'cross attenuate' control), but also by adjusting
  phase and delay across channels (the 'cross delay' control).  The
  A-B slider then controls how far the input apparently images toward
  the A output bank or the B output bank. 

  The mixdown blocks can also be configured to take as input not only
  the input channel, but also each channel's left (A) or right (B)
  reverb output.  For example, reverb left can be mixed to the left
  output, the reverb right mixed to the right channel and the original
  input crossplaced somewhere between left or right.  The input
  placement can be altered on the fly thus apparently moving the
  input's location while the impression of space the reverb creates
  holds still. For even more realism, adding an additional delay of 10
  ms or so to the reverb (sound travels roughly one meter in 3 ms) can
  pull the original image closer without losing the impression of
  reverberation in a medium-to-large hall.  A reverb delay also allows
  the use of a faster/tighter reverb time without losing the
  the impression of size.

  Of course, with eight channels, one can begin imaging/mastering for
  more than just stereo...

>>>> Don't we already have several free apps that do this sort of thing?

  The short/wrong answer is maybe. The complicated answer is no.

  .) I needed a specific set of filters
  .) I needed them in one place working together
  .) I needed to hear changes I made to settings as I made them
  .) I needed to be able to absolutely trust the filters would function
    as expected
  .) I needed it all to be convenient to use

  Given my specific requirements, nothing else came close to filling
  the niche and I didn't want to cobble together a 90% solution out of
  multiple other apps when this functionality was the very core of
  what I needed for mixdown.

  Both the Postfish UI and the filter functionality are intended to be
  the most usable day-to-day set, rather than sporting the maximum
  number of buttons in the smallest space or more features than the
  next app (or the slickest skin). 

  Postfish is the way it is because I need it, and I use it for the
  core basics of mixing that I absolutely cannot afford to screw up.
  Some filters (like the declipper and deverberator) are unique.  Even
  among those that aren't, Postfish as implemented deliberately sets
  speed/quality tradeoff much higher than most existing apps.

  The multiband compander is a case in point; other free apps do
  implement this effect. To my knowledge, all use the simplest/fastest
  method, operating directly on the FFT of a windowed block.  An
  FFT-based multicompander is fast, but the aliasing and frequency
  multiplication artifacts (you eventually end up multiplying the
  input by the transfer function of the window shape; most noticeably,
  they tend to cause odd pitch changes in raw vocals) render them
  unsuitable for professional-quality work.

>>>> What does the Postfish require?

  .) Linux 2.x (ports come later) with OSS or ALSA OSS emulation
  .) Libraries: FFTW3, pthreads, Gtk2
  .) Gcc and gmake
  .) A sound card or external USB/Firewire A/D/A
  .) A video card, preferrably one with fast AGP
  .) Alot of CPU.  Really.  As much as you can throw at it.  Dual Xeon 3GHZ? 
    Yup, you can use all of it.

  Seriously, this is a very CPU hungry app because of the
  aforementioned speed/quality tradeoff.  I can do simple mixdown of 8
  channels to stereo with a few effects in realtime on my G3-400, but
  the machine is crying.  The dual Athlon 2600 keeps up much better,
  but it's still possible to overwhelm it by lighting up all the
  effects and feeding it enough input channels.  The declipper,
  especially, eats CPU on heavily damaged audio.  The multiband
  compander is runner up in 'absurd levels of CPU usage'.

  Postfish, BTW, is designed to scale to dual CPUs.  A dual
  Athlon/Pentium/UltraSparc/PowerPC runs Postfish much better than a
  single processor.

>>>> Why is this a pre-release?

  Because it's not finished.  A few things are more obvious than others:

  1) The 'cue sheet' and 'settings list' panels are still
    unimplemented.  These are the only inactive features on the UI,
    but they're right on the main panel.

  2) Postfish *should* be a JACK-able app, but isn't.  That too should
    be done for a real release.

  3) Although the whole thing is designed as a rendering engine
    wrapped up in a neat async-safe library that's then used by an
    asynchronous UI, the source isn't entirely arranged that way.  It
    should be.  It will be for final release.

  4) The stereo reverb code used by Postfish is Steve Harris's GPL
    plate reverb from LADSPA.  Although this is a solid, reliable
    reverb implementation, it's also a bit thinner on final output
    results than I'd like.  Don't get me wrong, this reverb is
    excellent code, but it places emphasis more toward 'fast' rather
    than my desired 'best possible results given unlimited CPU'.  I'm
    evaluating other reverb implementations; this is a case where
    others have achieved clearly better results than I would, and
    thus I plan to use the best available to me.

  5) This code is just now seeing light of day.  It is probably *full*
    of simple bugs.  I'm confident in the audio pipeline itself (I've
    hammered on it mercilessly) but there's certainly many UI
    interconnection bugs left to find.

  6) Everyone knows a release requires documentation.  There is no
    documentation.

>>>> There's no documentation!?

  Not yet; good documentation requires effort and time. 

  That said, if read the list of effects, knew what they were, and
  knew basically how to use them, you should be able to pick up the
  Postfish and do useful work in a few minutes of playing around.

  I took care to establish and follow conventions in the UI: If you're
  using the shipped postfish-gtkrc theme, square blue buttons turn
  things on.  Triangular blue buttons pop configuration windows.

  The grid of buttons on the right in the 'channel' frame are the
  effects for the input channels.  The square blue buttons turn a
  specific effect on for a specific channel.  A triangle pops the
  configuration for that effect.  Mixing controls are labelled
  "Atten/Mix" in the lowest row in the "channel" panel.

  Further right is the "master" panel; these controls work the same
  way for effects on the output channels after mixdown.

  Finally, postfish -h will tell you how to get audio in and read
  audio out.

  Only two things are probably impossible to figure out just from
  an afternoon of playing around:

  1) Output from a reverb effect in the "channel" panel is sent to a
    separate internal stereo bus.  The reverb has to be explicitly
    mixed into the output on a 'mix' panel.  When postfish starts for
    the first time, the default setup mixes one side of the reverb
    pair back into left or right.

  2) Turning "Atten/Mix" off for any channel will mute it completely
    in all input effects, not just in the mixdown (this obviously
    doesn't affect output, but it also silences all the VU meters for
    that input channel, and that can be disconcerting if you don't
    expect it).

>>>> How do I get, build and install it?

  Postfish is in Xiph.Org's Subversion repository.  Get the source using:

    svn co http://svn.xiph.org/trunk/postfish postfish

  Edit the Makefile to select the proper 'ADD_DEF' line.  LinuxPPC
  users want the first, almost everyone else wants the second.  It
  should be self explanatory given text in the Makefile.

    cd postfish
    make

  and as root,

    make install

Happy hacking (and mixing),

Monty
TD, Xiph.Org

release of the Postfish

Reply #1
Wow, Monty, you've been busy! I don't even pretend to know how you could fit this in with all the other Xiph stuff. Pretty sweet though.

release of the Postfish

Reply #2
Quote
Wow, Monty, you've been busy! I don't even pretend to know how you could fit this in with all the other Xiph stuff. Pretty sweet though.

the Postfish has been a project I've been tinkering on for years.  This isn't exactly new software, just newly in a more releasable quality :-)

Monty

release of the Postfish

Reply #3
I wish windows-compatible version also!!!

release of the Postfish

Reply #4
Quote
I wish windows-compatible version also!!!

I remember in this post that QuantumKnot had compiled a postfish version. But I don't know exactly if it's a win32 or linux version...

release of the Postfish

Reply #5
Quote
I remember in this post that QuantumKnot had compiled a postfish version. But I don't know exactly if it's a win32 or linux version...

He only said he would investigate. He never reported success.

release of the Postfish

Reply #6
You'll probably need glib-dev, gtk-dev, and fftw3-dev to compile it. I'll tell you all the dependencies when I finish building..

edit:
It looks like it won't build with gtk 2.4.1.
outpanel.o(.text+0x371): In function `outpanel_state_from_config':
: undefined reference to `GTK_OPTION_MENU'
outpanel.o(.text+0x38d): In function `outpanel_state_from_config':
: undefined reference to `GTK_OPTION_MENU'
outpanel.o(.text+0x3a9): In function `outpanel_state_from_config':
: undefined reference to `GTK_OPTION_MENU'
outpanel.o(.text+0x3c8): In function `outpanel_state_from_config':
: undefined reference to `GTK_OPTION_MENU'
outpanel.o(.text+0x3e7): In function `outpanel_state_from_config':
: undefined reference to `GTK_OPTION_MENU'
outpanel.o(.text+0x406): more undefined references to `GTK_OPTION_MENU' follow
collect2: ld returned 1 exit status
make[1]: *** [target] Error 1
make[1]: Leaving directory `/home/kjoonlee/xiph/postfish'
make: *** [all] Error 2

edit2:
or maybe I'm doing something horribly wrong..

release of the Postfish

Reply #7
It looks like gtk 2.4.1 is missing the define for no reason.  It was in previous releases, and the functionality isn't marked deprecated in any online GTK docs...  2.0 through 2.3 have it....

Even weirder, the define *does* exist in the GTK2.4.1 source includes. Plain old Gtk2 bug?

[edit: I'm just an idiot.  See below]

Monty

release of the Postfish

Reply #8
Ah *ha*.  The docs still say to use it.... but in the *source*, gtk_option_menu is suddenly listed as 'deprecated' as of 2.4.  WTF!?

[edit: my docs say to use it because my docs were a few months old :-]

Monty

release of the Postfish

Reply #9
OK!  I have fixed the build for 2.4.1 for the moment.  Eventually I will have to replace the use of GtkOptionMenu in the output panels, but for the forseeable future, simply turning off the super-strict-no-deprecated-code defines allow things to build and function properly.

kjoonlee: you should be able to build now.

Monty

release of the Postfish

Reply #10
Yay! Success.

Put it somewhere in your path, then run something like:
postfish The_Corrs-I_Know_My_Love.wav > /dev/dsp

(edit: Not the best way to run it if you want to save results.)

You can run postfish -h to learn more.

Here's the ldd output:
Quote
libgtk-x11-2.0.so.0
libgdk-x11-2.0.so.0
libatk-1.0.so.0
libgdk_pixbuf-2.0.so.0
libm.so.6
libpangoxft-1.0.so.0
libpangox-1.0.so.0
libpango-1.0.so.0
libgobject-2.0.so.0
libgmodule-2.0.so.0
libdl.so.2
libglib-2.0.so.0
libpthread.so.0
libfftw3f.so.3
libc.so.6
libX11.so.6
libXrandr.so.2
libXi.so.6
libXext.so.6
libXft.so.2
libfreetype.so.6
libz.so.2
libfontconfig.so.1
libXcursor.so.1
libXrender.so.1
/lib/ld-linux.so.2
libpangoft2-1.0.so.0
libexpat.so.2

release of the Postfish

Reply #11
I still think this deserves an announcement in the News Submissions (which I made just now but realised instead it was in here) since it wasn't that obvious Monty would announce it in the Digital A/V News forum and it is an audio tool after all.

OK, time to checkout

release of the Postfish

Reply #12
I've tried compiling the Postfish.  But I got this compile error:

Code: [Select]
windowbutton.o(.text+0x157): In function `windowbutton_draw_indicator':
: undefined reference to `_gtk_check_button_get_props'
collect2: ld returned 1 exit status
make[1]: *** [target] Error 1
make[1]: Leaving directory `/home/stephen/xiphsvn/postfish'
make: *** [all] Error 2
Exit 2


I'm using GTK 2.4.0.

release of the Postfish

Reply #13
Quote
I've tried compiling the Postfish.  But I got this compile error:

Code: [Select]
windowbutton.o(.text+0x157): In function `windowbutton_draw_indicator':
: undefined reference to `_gtk_check_button_get_props'
collect2: ld returned 1 exit status
make[1]: *** [target] Error 1
make[1]: Leaving directory `/home/stephen/xiphsvn/postfish'
make: *** [all] Error 2
Exit 2


I'm using GTK 2.4.0.

Thanks for the report...

This is an internal function in the gtk_check_button class I'm subclassing from.  It exists in Gtk 2.2 and 2.4.1, but is missing in 2.4.0.  So a quick gtk upgrade will correct this.

However, looking at this strategically it's obviously an unsafe call to be relying upon (if the leading underscore didn't give that away ;-).  The good news is it's just a small wrapper around a few supported/exported functions, so I'll move the wrapper up into windowbutton.c

Monty

release of the Postfish

Reply #14
This was quick a fix, and I committed it.

Monty

release of the Postfish

Reply #15
Excellent.  It now builds perfectly in 2.4.0 now.  But then, I downloaded 2.4.1 for nothing

Anyway, thanks Monty

release of the Postfish

Reply #16
hmm....Here is a snapshot of it.  See the triangular buttons.  Looks like they've been offset a bit.  Cosmetic issue I guess

Postfish Screenshot

Looks very slick indeed 


EDIT:  Changed image to URL since the pic is a bit big.

release of the Postfish

Reply #17
Quote
hmm....Here is a snapshot of it.  See the triangular buttons.  Looks like they've been offset a bit.  Cosmetic issue I guess :)

Looks very slick indeed  :cool:

Not sure what you mean by 'offset' as the pic you posted won't come up here.  If you mean 'the window poppers overlap the activity buttons', that's intentional.  When you have 16 input channels, width becomes an issue, thus the horizontal compression of the Channel and Master panels.

[edit: Whoops!  There it is! :-)  Yes, that's all as its supposed to be.  Well, except the monospace font for number readouts is supposed to be 6x13 or smaller for optimal formatting... but libpango makes gettting the desired monospace font really hard]

Monty

release of the Postfish

Reply #18
Oh I understand.  Looks like rarewares.org is having problems again.

Another point for people who might be interested in is that they should compile the single precision part of fftw.  My existing  fftw was compiled in double precision (since I needed that for SSE2) and the Makefile couldn't find fftwf-wisdom.  So I made a single precision compile of fftw as well

release of the Postfish

Reply #19
Quote
Oh I understand.  Looks like rarewares.org is having problems again.

Another point for people who might be interested in is that they should compile the single precision part of fftw.  My existing  fftw was compiled in double precision (since I needed that for SSE2) and the Makefile couldn't find fftwf-wisdom.  So I made a single precision compile of fftw as well :)

Yes, single precision fftw is required.  BTW, I had terrible trouble with FFTW3 compiled with SSE support: mostly it crashes without doing anything useful (and yes, my processor has SSE support).  This is an issue with FFTW3, not Postfish and I can't seem to do much about it; I had to use the stock debian FFTW3 without vector unit support.

Also, I don't know how Linux sets up SSE and what default exception mode it uses.  I know that Linux sets up Altivec in 'useless' mode so that any denormalized math just crashes the program.  *That* I found an ugly but working fix for (see the top of main.c), so Altivec is OK.  Stupid thing is that GCC's C99 support is supposed to be able to set this all up through fedisableexcept() except GCC only implements the functionality for the core FPU and not for the vector unit (ie, Altivec or SSE) so there's no portable way to put the vector unit in a known, sane state on Linux.

If you get your FFTW3 with SSE to work properly... or have an answer to any of those unkowns... let me know :-)

Monty

release of the Postfish

Reply #20
Quote
Oh I understand.  Looks like rarewares.org is having problems again.

Another point for people who might be interested in is that they should compile the single precision part of fftw.  My existing  fftw was compiled in double precision (since I needed that for SSE2) and the Makefile couldn't find fftwf-wisdom.  So I made a single precision compile of fftw as well :)

Yes, single precision fftw is required (debian at least installs both single and double precision version).  BTW, I had terrible trouble with FFTW3 compiled with SSE support: mostly it crashes without doing anything useful (and yes, my processor has SSE support).  This is an issue with FFTW3, not Postfish and I can't seem to do much about it; I had to use the stock debian FFTW3 without vector unit support.  It appeared to be a simple logic problem in FFTW3 plan generation; the SSE itself is fine, the segfaults happen before that.

FFTW3 with altivec is fine, and Postfish does properly handle Altivec exception setup via some ugly but workable gcc-intrinsics-based fixes (see the top of main.c), so Altivec is OK.  Stupid thing is that GCC's C99 support is supposed to be able to set this all up through fedisableexcept() except GCC only implements the functionality for the core FPU and not for the vector unit (ie, Altivec or SSE) so there's no portable way to put the vector unit in a known, sane state on Linux.

If you get your FFTW3 with SSE to work properly... let me know :-)  I do know that Linux does set up SSE to mask exceptions by default (unlike Altivec), so that part is fine.

Monty

release of the Postfish

Reply #21
The only word that comes to mind, regarding the interface, as I play with this is: slick The spinning fish and the smooth graphical bars moving....very nice (and slick)  . 

I opened up the equalizer as I was playing a file and tried to move the sliders but I couldn't move any.  They seem to be stuck.  Same for any other slider control.  They highlight as I move my mouse over them, but I can't seem to drag them.

release of the Postfish

Reply #22
Quote
I opened up the equalizer as I was playing a file and tried to move the sliders but I couldn't move any.  They seem to be stuck.  Same for any other slider control.  They highlight as I move my mouse over them, but I can't seem to drag them.

Perhaps you need that gtk 2.4.1 upgrade after all? ;-)

Actually, when you try to click/drag, is the focus moved?  It appears mouse button presses are disappearing.  That's weird; the widget uses very correct event hooks and is only aware of what GTK tells it....  The fact that the prelight appears means it's getting mouse enter, so it's odd that button press would get swallowed.

Monty

release of the Postfish

Reply #23
I guess that's the story of my life:  things that work for most never work for moi

Here is the exact behaviour I get.  I move my mouse over a slider and it turns light blue.  When I click on the slider, it goes dark.  But the moment I drag it, it lights up to light blue again.

I think it's time to upgrade to 2.4.1 (might as well since I downloaded the source).  Things can only get better (or I hope so anyway).

release of the Postfish

Reply #24
On my side, it looks like modifying the EQ settings in real time takes up huge amounts of CPU cycles. It's easy to move the sliders when playback has been paused, but it takes patience to move the sliders in "real time."

edit: Oops, mistake on my side. It isn't so slow when I'm just modifying the EQ.

 
SimplePortal 1.0.0 RC1 © 2008-2019