[linux-audio-user] agc, long time event recording (bounty)

hi there,

hope this mail is on topic on this list.

for an FM and internet radio project during the
worldcup 2006 in Berlin, we're looking for support on
the linux audio developer side. it's a cultural project
and we'll recording audio events at different
places, from clubs to universitities. there's a small budget
included and some preliminary code is available from
an older streaming project.
of course the project will be released under GNU.
if you are interested to join in and help us with your
work, please contact us quickly at:
kontakt->radioeinszueins.de

here's the basic setup we're working on,
it derives from years of practical d.i.y.
experiences with recordings and streaming
of club events.

your questions and comments are welcome.

.....................................................................................................................
>> automatic event recorder:

questions/tasks:
- how to profit from 24bit to 16bit dithering and 88.2 khz to 44.1 kz resampling
for dynamic limiting/compression aka "Automatic Gain Control"?
- silence detection (pause of recording or auto cutting)
- premastering to have recordings broadcast-ready and normalized
(up to the much hated brickwalled optimod FM sound..)
- tradeoff between sound quality and throughput
- presets for speech, music (jazz, classic, rock, electro), room micros
+ making podcast "secure"
+ usability, maintainablity, error control + watchdogs.

hardware: terratec phase 22, amd 1400 mhz,
512 RAM, 160 HDD, 2HE case connected to lan/dsl
router and balanced audio signal from mixer / pa.

input:
scheduler data via ical and/or dublin core metadata.
analog balanced stereo audio in, 0 - 17 db

output
podcast xml, mp3 lame 128kbps
archive quality luxury version: ogg, flac
including appropriate id3 tags with event metadata.
file repository (like apache-modmp3) with secure access
logs and error reports

control:
via web interface

proposal:
when you do not have the time to master large amounts
of recorded audio material by hand, an automatic gain control
at the time of the recording could help a lot.
aiming at a good tradeoff between dynamic compression and
sound quality, the main issue is to get a good leveled signal
in the digital domain. it could be done by auto-readjusting the input
gain, or by using using the headroom of 8 bit,
before the 24bit to 16 bit conversion for some smart compressor/limiter
magic in real time. the mechanism of dithering/downsampling is known from
mastering at the end of the chain (e.g. the waves l-1 maximizer).
so why not using it for an unmaintained non-annoying AGC?

maybe appropriate algorithms are available from voip projects,
or by directly using vst plugins, or tuning of the jamin multiband
compressor. a window manager like gnome is not obligatory.
we're experimenting with a plugin-chain for
alsa, jackd, ecasound based on ladspa/fst.
by now we can get a basic setup, but we're not even
sure if the AGC is realizeable in this short time.. some
basic code is available but we'd need help with
making it really run on a more advanced level.
Received on Thu Apr 13 04:15:09 2006