21 Sep
2022
21 Sep
'22
4:03 a.m.
Hi!
Don't know, if this is actually the academic or scientifically correct
way to do it, but I got good sounding results - in fact, I managed to
insert mono-recorded voices into an ambisonic recording with other
people and it didn't sound that weird…
But I'm always interested in getting things more and more perfect, so I
hope many of the ambisonics-pros are joining this thread to discuss the
topic.
This is the software and hardware I used:
* Røde NT-SF1 1st order Ambisonics microphone
* Sound Devices MixPre-6 recording device
* Ubuntu Studio 04.22
* Carla 2.4.2
* Røde Soundfield PlugIn (Windows 64bit VST3)
* Wine
* Ardour 6.9
* Audacity 2.4.2
* LSP-PlugIn Suite (LV2)
* IEM Ambisonics PlugIn Suite (LinuxVST)
And here is the whole procedure, how I made it…
I made an ambisonics recording with people in a room. After that I took
a paper bag to create an impulse, about 50cm away from the NT-SF1. (Just
clapping the hand was too silent in my opinion and I didn't want to use
balloons because of the plastic waste. Paper bags seem to be an
relatively environment friendly and cheap alternative to me.) I retried
as often as it needed to get a good sounding and non distorted record of
the impulse and its respond.
Getting back to home I copied the A-format of my recordings to my
computer, fired up Ardour, created a 4-channel track and inserted the
16-channel-version of carla patchbay into that channel strip. I had to
deactivate the panner in every channel strip to make Ardour "ambisonics
compatible", yes, also in the "Master"-channel which must be blown up
to 4 channels, too. I imported the tracks that I wanted to be converted.
The only tool in my procedure that wasn't free and open was Rødes
Soundfield PlugIn, so I had to use some tricks to get manufacturer
proven A-to-B conversion of my recordings. FalkTX did make carla
windows-VST-capable, but it's not a bullet prove task… So, I got the
Soundfield exe-file, installed it with wine and I finally got a dll I
could drag and drop into carla patchbay's GUI. There it is! Chances are
Ardour crashes, so I made sure to save after every step. As input
format, I have to chose "NT-SF1", on output the best choice is to use
"B-Format (Ambix)", because Ambix is the actual standard and the IEM
plugins only handles Ambix. Of course, I had to connect Carla's output
to the input of Soundfield and vice versa.
Time for the A-to-B-convertion. Drag and drop the 4-channel-audio into
the editor window of Ardour, set "Start" and "End" and use export.
Make
sure to have all 4 channels to be exported into one file. You can use
both WAV or FLAC. Next step is Audacity. (You don't need to close
Ardour, sooner or later it will crash… :) )
Instead of stereo files, Audacity handles 4-channel-audio as 4 seperate
mono files. (But it can be exported to 4-channel-audio, again, if you
choose "advanced mix options" in the "import/export" division of the
preferences. You won't need it for the virtual ambisonics IR reverb.)
But - it is mandatory to keep the audio sample accurate between all the
tracks! If you want to delete, delete from all of them, simultaneously!
So, you choose your favorite BANG! out of the inpulse series you
recorded and delete the rest. Export every mono track as WAV. Use Ambix
nomenclature: 1st track is W, 2nd is Y, 3rd is Z and 4th is X. Best bet
is to write the number and the letter in the name of the file. I choose:
[nameoftheplace]_1_(W).wav and so on…
Now, to actually build the virtual ambisonics IR reverb, fire up Ardour
and make it ambisonics compatible (deactivate panner). Create a
4-channel-audio-bus and place a 16-channel Carla patchbay. In the Carla
patchbay, create 4 "LSP Impulse Responses Mono" - one for each channel.
(Using LV2 in Carla works pretty stable!) Open the GUI for each "LSP
Impulse Responses Mono" and use the corresponding WAV in it by clicking
into the GUI and choose the right file. Inside Carla patchbay it should
look as follows:
Carla channel 1 output ---> LSP Impulse Responses Mono #1 (with
[nameoftheplace]_1_(W).wav) ---> Carla channel 1 input
Carla channel 2 output ---> LSP Impulse Responses Mono #2 (with
[nameoftheplace]_2_(Y).wav) ---> Carla channel 2 input
Carla channel 3 output ---> LSP Impulse Responses Mono #3 (with
[nameoftheplace]_3_(Z).wav) ---> Carla channel 3 input
Carla channel 4 output ---> LSP Impulse Responses Mono #4 (with
[nameoftheplace]_4_(X).wav) ---> Carla channel 4 input
Make sure, that every LSP IR Mono instance uses exactly the same values
of headcut, tailcut and amplification. WIth headcut, you should cut the
impulse. Now, you can use a mono audio file, place it into the
ambisonics sphere f.e. with the IEM StereoEncoder, use an auxiliary send
(don't forget to disable the panner of it, too!) and enjoy the
reverberation! (Play with Ardour's "strict E/A" to get as much channels
for each strip you need!) Using an EQ after the virtual reverb is always
a good idea. LSP IR Mono has one integrated, but my impulses seem to
have no linear frequency band and needed more than the +/-12 dB the
plugin offers… (I take IEMs "MultiEQ" for this task - one setting for
all the channels is another advantage.)
Unfortunately, Carla can't recall LSP IR Mono's WAV-file, so, you have
to reload it, every time you start your session.
Greets!
I hope, this can be useful for anyone…
Mitsch
21 Sep
21 Sep
1:27 p.m.
AH!
I forgot a little info: You won't hear anything from "LSP IR Mono" as
long as you won't choose a "source" in the middle buttom of the GUI. I
choose "left" for all four channels - I don't know, what this is all
about, but as I already wrote: it won't work, without.
Also, I don't know, if "LSP IR Mono" will compensate, anyway, but
theoretically you should add the time to "pre delay" that was cut off
with "headcut". So, f.e., if your IR file starts accurately with the
impulse and you use 20 ms of headcut, I had good results adding 20 ms to
"pre delay", too.
Greets!
Mitsch
5:35 p.m.
On Wed, Sep 21, 2022 at 02:02:58AM +0200, Michael Jarosch wrote:
After that I took a
paper bag to create an impulse, about 50cm away from the NT-SF1.
If the room is big enough, you could use a larger distance.
This will reduce the amplitude of the the impulse but not
the room response, so you could have a better S/N ratio.
A paper bag (or balloon) is not ideal, there are alternatives,
see below.
The only tool in my procedure that wasn't free and
open was Rødes Soundfield
PlugIn, so I had to use some tricks to get manufacturer proven A-to-B
conversion of my recordings.
There is NT1 preset available for tetraproc (real-time A/B processor, Jack)
and tetrafile (file processor doing the same). I can send it if you want.
The method of choice today is not to use balloons or pistols, but
a logarithmic sweep signal and deconvolution with the inverse sweep.
This will give you
1. A more correct result, since the spectrum of the test signal is
exactly known (apart from the speaker used to play it),
2. A much better S/N ratio since the test signal can be several
seconds long so it has much more energy at the same amplitude as a
popping balloon.
Ciao,
--
FA
11:10 p.m.
Am 21.09.22 um 15:35 schrieb Fons Adriaensen:
If the room is big enough, you could use a larger
distance.
This will reduce the amplitude of the the impulse but not
the room response, so you could have a better S/N ratio.
Good hint! I always
thought it was necessary to place the impulse source
as close as possible to the microphone.
There is NT1 preset available for tetraproc (real-time
A/B processor,
Jack)
and tetrafile (file processor doing the same). I can send it if you want.
Yes! I
really missed using the commandline for A-/B-convertion. I'd
really appreciate that!
The method of choice today is not to use balloons or
pistols, but
a logarithmic sweep signal and deconvolution with the inverse sweep.
This will give you
1. A more correct result, since the spectrum of the test signal is
exactly known (apart from the speaker used to play it),
2. A much better S/N ratio since the test signal can be several
seconds long so it has much more energy at the same amplitude as a
popping balloon.
I know this procedure from speaker development. But I wonder which tool
is used to do the deconvolution in this scenario. What do you use?
The downside is that a good speaker with a wide frequency range is
certainly much heavier and bigger than my package of paper bags. :D (Let
alone the fact that amplification and signal generation could afford two
additional devices - and an additional power source.) I already thought
in the same direction, but I dropped it in favor of mobility.
Although, I already realised, that my paper bag impulse is not the best
option for audio quality and partly needs heavy postprocessing…
But I managed to accelerate finding the correct equalization setup, at
least:
Using the right head-/tailcut parameter, it's possible to isolate just
the impulse in the "LSP IR Mono" plugin. Stimulated by a pink noise
signal one can spectrum analyse the output and find out which
frequencies are underrepresented in the impulse. With IEMs "MultiEQ" one
can fill the gaps. Still - as long as you're trying to combine an
ambisonics recording with an afterwords integrated mono signal, f.e. -
it's necessary to use hearing and comparing with original recordings.
Altogether, a not-so easy and quick task. Besides the fact, that heavy
equalizing won't improve S/N ratio, at all.
Greets!
Mitsch
582
days inactive
582
days old
linux-audio-user@lists.linuxaudio.org
3 comments
2 participants
participants (2)
-
Fons Adriaensen -
Michael Jarosch