[Jackit-devel] Re: [linux-audio-user] ices-jack pico-howto

[Joern Nettingsmeier]

hi eric!


Eric Dantan Rzewnicki wrote:

> I just read through this again and a few questions popped up:
> 
> ices runs on the sound source and connects to jackd, correct? (I think
> this is what you state below, but just want to double check.)

yes. there is a jack graph that is producing sounds. ices-jack is 
just another jack client. it does the encoding to ogg and optional 
resampling and then sends the encoded stream to an icecast server 
(which can be on another machine or local, the transmission between 
source and server is via http)

> Have you, or has anyone you know of run jackd and ices on a box with no
> soundcard using the jackd dummy driver? (more on why below.)

yes, i did that during testing of the lac streaming setup this year. 
works nicely.

> If I want to stream out ogg vorbis, at which point does the encoding
> happen, on the audio source/ices/jack box or the icecast box?

in the ices process. so if you use two boxen, the big iron should be 
the jackd/ices box, since the encoding is computationally expensive. 
the streaming can be done by a leftover - i guess a pentium 233 
running icecast can saturate a 100mbit link. you want some memory 
though.

> Other than ices and icecast, do both the icecast box and the ices box 
> need all of the svn packages you list above?

the icecast box does not need the *-tools. (nor does ices strictly 
speaking, but they are nice to have). i described a "bleeding edge" 
best-of-xiph.org setup. if you don't want to play with video or 
very-low-bandwidth speech streaming, you can omit the speex and 
theora packages as well as flac. ices/icecast check for them at 
compile time - i wanted all bells and whistles, but if the libs are 
not present, those features will be left out with no problems.

you may even get away with using your distro's libvorbis, although i 
think ices-kh does require ogg2 and won't compile with plain old ogg.

>>now fire up icecast, fire up ices, connect it to your jack graph, 
>>and the fun starts.
>>the default config files are extensively commented, but here's my 
>>config, in case you need some more inspiration:
>>http://spunk.dnsalias.org/download/ices.xml
>>http://spunk.dnsalias.org/download/icecast.xml
>>(the source and server run on different hosts, and icecast runs 
>>chrooted and as user icecast)
> 
> 
> here source = ices, server = icecast? (again, just double checking.)

yes, sorry for my sloppy choice of words.

> I've planned a streaming self-feeding composition project to 
> specifically make use of the repeated vorbis encode/decode artifact 
> buildup. 

when you do this, consider that under ideal circumstances, re-ogging 
a pcm stream that has been ogg-encoded before will not add new 
encoding artifacts, unless you reduce the bitrate.

the degradation you hear on my recording is mostly due to repeated 
ad/da conversion, a mixer that catches harddisk hum and a reduction 
of the level....

maybe ogg->wav->mp3->wav and back would be more interesting. both 
codecs should have different data reduction models and ideally not 
only take away information but create artifacts different from each 
other.

or maybe add an analog edge to your graph (pun intended :-).

> The basic structure will have:
> 1) a single original source audio file;
> 2) several instances of a script that selects a random section of the
>    audio file, either stretches or squishes the section's time scale,
>    possibly reverses it, and then loops and plays it out to jack ports
>    with various randomly parameterized controllers controlling aspects 
>    of the sound (panning, volume, etc.);
> 3) ices connected to jack and streaming the output of the several
>    instances of 2) to icecast;

you need one ices process per such instance. ices can encode to 
different bitrates, but handles only *one* incoming stream (which 
can have an arbitrary number of channels).

> 4) recording script that records from jack ports and writes ogg vorbis
>    files to a pool from which the 2)'s will select on subsequent 
>    iterations.

hmm. the internet tranmission does not generate any interesting 
additions, except for possibly varying round-trip times or maybe 
dropouts....

> The script in 2) will have to decode the ogg files before selecting a
> chunk and manipulating it. I guess since jack only deals with floats 3)
> and 4) will have to encode to vorbis separately, unless 3) can be
> scheduled to start a new dump file, say, every hour.

i don't understand this, but yes, ices can be made to start a new 
dump file at any time by sending it a USR1 signal. make it a cronjob 
and there you go.

> The whole thing will be given a disk space quota. Probably 4) will check
> this and choose previous output files at random to delete in order to
> stay within the quota.

:)

> I plan to do 2) & 4) in ecasound python eci scripts. Each instance of 2)
> will, hopefully, run continuously selecting the duration of it's runtime
> on each iteration and simply starting over by selecting a new section of 
> audio from the pool of files when it finishes. 4) will probably be 
> scheduled hourly via cron. Hopefully I can get this thing to run 
> continuously for years on end. Since it won't actually be sending any 
> audio out to a soundcard I would like to run it on my fileserver with 
> jackd using the dummy driver.
> 
> Anyway, that's the plan. We'll see if I ever get it implemented ...

good luck!



-- 
"Debugging is twice as hard as writing the code in the first place.
Therefore, if you write the code as cleverly as possible, you are,
by definition, not smart enough to debug it."
	- Brian W. Kernighan

Jörn Nettingsmeier
Lortzingstr. 11, 45128 Essen, Germany
http://spunk.dnsalias.org (my server)
http://www.linuxaudiodev.org (Linux Audio Developers)