[Clam-devel] SampleRate problem in chordata (and NetworkEditor) related with the PA backend (problem and proposed solutions)

David García Garzón dgarcia at iua.upf.edu
Fri Feb 27 05:11:17 PST 2009


A Divendres 27 Febrer 2009 06:05:51, Hernán Ordiales va escriure:
> I think this is a very well known bug, i'm talking about the PortAudio
> audio backend sampleRate value hardcoded to 48000Hz[1]. I always been
> using jack as backend for NetworkEditor, so i just realized about this
> bug playing with chordata, which only has PortAudio as audio backend.
> By doubts, the problem seems that playing a file at 44100Hz, if the
> backend is set-up at 48000 there is not automatic sampleRate
> conversion and you get a distorted pitch.
>
> Having in mind most common audio files are recorded(saved) in 44100,
> the audio/feedback result in chordata is specially awful. So i believe
> this should be solved ASAP, at least for chordata, since is more
> common end user oriented.
>
> After found this problem, i tried many 'solutions':
> * I recompiled the portaudio backend with 44100 as default sampleRate
> and all ok.
> * I replaced portaudio with jack as backend in chordata since with it
> it's feasible to set-up the SR of the backend when you start the jackd
> daemon (but this solution excludes platforms differents from linux)
> * Resample before the output if the SR of the input file does not meet
> with the current backend SR (in the case of portaudio)[2]. The problem
> here is that increases the cpu load and seems more vulnerable to
> underflows.
>
> Other solutions:
> * Set up the PortAudio backend depending on the file that's going to be
> played? * ?
>
> what do you think? which is the best way to overcome with this issue?

I would be very glad if you can manage this long term pending issue. The ideal 
setup for me I would be wave file handlers do the transcoding to a given 
sample rate nominal to the network (as the network could have samplerate 
dependencies) and then the backends transparently recode network samplerate to 
the one of the device (or trying to configure backend to the network 
samplerate). Devil is on the details so maybe we should write down some use 
cases to think on each solution whether it fits, or at least writing down 
current ones in order not to generalize too much.

But whatever the solution we choose what we need is something to resample and 
current resample plugin does NOT work. So if you based your new processing in 
the old one, i guess they wont work either.

The main problem is that I was using the 'feed-the-whole-audio' simple 
interface. This is not ok for most ratios where the processing just fails to 
output anything.

Libsamplerate offers an streaming interface where you feed an input and an 
output buffer but as far as i read it doesn't give you any guaranties that it 
will fill the output buffer for a given input size. I guess we should look to 
the source code to know which are the needed framesizes in order for us to set 
up the latency. I got blocked there and stopped the spike so if you want to go 
deeper on the research, it is perfect.


> Related with this, recently i faced with another problem that i've
> been _suffering_ lately, NetworkEditor is always loading jack as
> backend, beyond jackd is running or not (for me it's a problem because
> sometimes i'd like to run NE with PA as backend). Checking the code i
> learned there is a backend autoselection process, first tries with
> jack (if it's present), then with PA. I'd like to have the option of
> let me force the backend. Related to this, I did an small proof of
> concept using command line[3], but, the best _framework_ [4] (i mean
> by this an object oriented solution, free as speech, small, easy and
> clear) i have found to parse command line options is qt based. Just as
> an example, looks like:
>
> 	GetOpt opts(argc, argv);
> 	bool help;
> 	opts.addSwitch("help", &help);
> 	QString backend;
> 	opts.addOption('b', "backend", &backend);
> 	QString file;
> 	opts.addOptionalArgument("file", &file);
>
> Very similar to scons options, don't? but I don't fully like the idea
> about include an external framework (altough it's just a couple of
> files) for command line parsing based in qt because it couldn't be
> cleanly reused in command line apps (where a lot of arguments have
> more sense!)
>
> Ok, we still can implement it in a simpler way[5], although OTOH,
> overload a graphic application with command line options, maybe isn't
> the best way to go, but NE is missing of a configuration
> section/window. What do you think about this?

Code for that is in the Prototyper which allows to provide a sorted list of 
preferences. I planned to unify NetEditor with that code in Prototyper. Anyway 
if you change the NetworkEditor options think on Prototyper too.

About introducing more libraries, please be carefull so that the libraries are 
available at least as package in debian/ubuntu and can be (cross)compilable 
for windows and mac. Introducing a non-portable dependency in such a central 
part of the framework could mean losing support for a platform.

Also about the options we are considering to introduce boost.program_option[1] 
which is available in most platforms and we already have used it in the 
raytracer.

[1] http://www.boost.org/doc/libs/1_38_0/doc/html/program_options.html

If the library you suggest were integrated into Qt, it would be perfect as we 
wouldnt need to include a new library but i guess is a third party based on 
Qt, not just Qt. Also if it is about two files and the license is compatible 
we could include it. Hey, i meet people from froglogic at the QtDevDays.

> Well, just let me know if you find some of these hacks useful and
> would you like to see them commited (or check the patchs) or i just
> keep them as my own workarounds to solve my own needs... :P
>
>
> Cheers
>
> [1] CLAM/src/Flow/Networks/BackEnds/PANetworkPlayer.cxx
> PANetworkPlayer::PANetworkPlayer()
>
> 	: mPreferredBufferSize(paFramesPerBufferUnspecified)
>
> 	, mSamplingRate(48000)
> 	, mPortAudioStream(0)
> 	, mError(paNoError)
>
> why?

1. No resampling algorithms integrated so we need to supose on SR for all
2. Most modern soundcards are 48k
3. We have some requirements of 48k for barcelonamedia
4. We moved most algorithms/backends to default 48k


> [3] NetworkEditor --help
>  Usage: NetworkEditor [file]  Network filename.
>                       [-b OR --backend]  The backend can be JACK,
> PortAudio or Auto.
>                       [--help]: This help.
>
> ./NetworkEditor -b PortAudio
> ./NetworkEditor -b JACK
>
>
> [4] I tested:
>            * getopt (ported to qt4 by me)
> http://www.froglogic.com/pg?id=PublicationsFreeware&category=getopt
>            * http://code.google.com/p/simpleopt/
>            * cmdLine, argtable, etc
>
> [5] please understand i'm lazy, you'll agree with me that is more
> funny to code audio dsp related things

-- 
David García Garzón
(Work) dgarcia at iua dot upf anotherdot es
http://www.iua.upf.edu/~dgarcia





More information about the clam-devel mailing list