Re: [Linuxsampler-devel] 24bit multi-io audio card compatibility, Was: CVS

[<be...@ga...>]

crive Mark Knecht <mar...@co...>:

>
> Benno,
>    OK, I came home for lunch since I was curious about trying this out.
> I am getting sound from LinuxSampler, so congrats. That's great!
>
>    I only have one GSt library on this machine right now, which is the
> 'String Boxes' library from Sonic Implants. I have loaded the 8 voice
> Mellotron Choir gig file. I am listening to both GSt and LSt them side
> by side. I am able to get both single notes and polyphony from LSt.
> That's good. Some data:
>
> 1) The first issue is that the Mellotron voice I'm playing is 10 seconds
> long on GSt, at which point I hear it loop and start over. (Hit one key
> and hold it for 30 seconds and you hear 3 passes.) However, LSt is only
> playing the voice for 2 seconds and then quitting. It doesn't loop the
> voice after 10 seconds, it just stops.

Keep in mind that looping is not implemented yes.
If LS is playing only for 2secs then it is playing a sample that is only.
2secs long. The wrong pitch is because LS is not honoring all the articulation
parameters yet. Christian will at that soon.
This will correct all errors you encountered.
(wrong sample triggered, thus often it sounds out of tune or the note
scale is completely screwed up).

For example if you try the 100MB church organ from G-Town:
http://62.13.11.115/gtown/

it sounds bad too because the note scale does not sound correctly.


>
> 2) The pitch of the LSt voice doesn't sound quite right. It seems too
> high and brittle. Twice as high, possibly? This is most likely that
> we're not picking up the intended middle C correctly I think. (C3 or C4
> - it changes from file to file I think.)

This is normal for the reason stated above.
Christian hurry up ! :-)
>
> 3) Zooming way in while in Pro Tools shows the latency of LSt to be
> slower than GSt. It doesn't look horrible, and I didn't try to measure
> the difference as I don't have much time right now. We can focus on that
> later.

Hmm this is interesting. How do you measure the latency ?
Can you record both the time a midi event gets in and put that on the
same time scale as the audio in PT ?

IIRC you played LS with --fragmentsize 256, lower this to 64
it should give you a latency of 2*64 = 128 frames = 3msec

Please redo the test and tell us if the latency is correct
(probably you need to add the 1.1msec of MIDI Note-on latency if you trigger
LS via external MIDI device).

>
>    I set up a quick Pro Tools session and recorded them both at the same
> time from the same key hit. The 24-bit zipped file with both sounds
> separate is about 2MB. That is probably too large for me to send to the
> reflector. Shall I send you a copy directly? I'm attaching the shorter,
> 16-bit version here, but at 350K your reflector may bounce it. Do you
> have an ftp site somewhere where I could drop audio like this? Maybe
> even in CVS somewhere?

I could open you an account on the LS.
But I'd like to wait for Christian's comments if it is too early
for such tests. (because he must add articulation otherwise all but the
simple GIGs will play incorrectly).

> Certainly. I didn't take it the least bit personally. Please don't take
> offense in the future when 100's of people coming from years of GSt
> background ask the same questions. Take some time and write a position
> statement on it today, put it on the web site and let's just point
> people towards it. The only thing I can promise you is that I'm not
> going to be the last to ask it!! ;-)

You are completely right.
As long as users are doing us an excellent service like you are doing
 they are allowed to pose this question :-)

PS: your message with attachment probably did not make it through
the mailing list (I'll forward the file to Christian).
In future please keep the test files on your box and we will agree
off-list where to put them.

PS: the current version of LS simply sums up voices and divides the
result by 4:
in audiothread.cpp:
sample_point = this->pAudioSumBuffer[u] / 4; // FIXME division by 4 just for
testing purposes (to give a bit of head room when mixing multiple voices together)

this means your headroom is about 12db (1/4) so don't be surprised if you
hear clipping when lots of voices are playing.

We will make this configurable or some good soul will add a limiter/compressor.
But I think the best solution would (when jackd support will be in) to
simply output the audio in 32bit float and then run it through external
 compressor/limiter/amp plugins.


cheers,
Benno
http://www.linuxsampler.org



-------------------------------------------------
This mail sent through http://www.gardena.net