Linux Vocoders

[danboid]

Today I felt the need to 'do a Cher' so I tried following Autostatic's qtractor vocoder video:

http://www.youtube.com/watch?v=tusCeI1aQ4c

I used the same plugin but I was trying to do it with the JACK standalone version of Aspect instead of Yoshimi, not that it should really matter as long as its routed correctly. I'm hoping to get this working nicely on my 1.6Ghz atom netbook which is obviously more convenient to lug to gigs that my laptop so thats what all I've tried it on so far but I just get xrun mania and I have to kill -9 qtractor. I've not tested my suspicion but I suspect Aspect is lighter on CPU than Yoshimi - its certainly a lot less taxing than Noisemaker anyway so I thought that may help lighten the load a bit and I really don't like Yosh/Zyn - not their clunky, oversized and netbook-hostile GUIs anyway!

Note that I was using my Focusrite 2i4 for mic input hence it would've been having to deal with a 32 (24) bit input - maybe I'll have to force 16-bit audio to vocode on something as puny as my netbook?

Apart from the Sirlab vocoder I tried today, there is also the MDA and TAL vocoder plugins and apparently Rakarrack has a vocoder too. Has anyone tried all four? Which is the most CPU efficient? Which sounds the best? Which gives the best quality vs CPU usage? Any other vocoding tips you have to share?

Thanks for sparing me the experimentation!

[autostatic]

I used the same plugin but I was trying to do it with the JACK standalone version of Aspect instead of Yoshimi, not that it should really matter as long as its routed correctly. I'm hoping to get this working nicely on my 1.6Ghz atom netbook which is obviously more convenient to lug to gigs that my laptop so thats what all I've tried it on so far but I just get xrun mania and I have to kill -9 qtractor.

That's weird, I've used this setup successfully on a 1,2 Ghz netbook (AMD L110).

I've not tested my suspicion but I suspect Aspect is lighter on CPU than Yoshimi - its certainly a lot less taxing than Noisemaker anyway so I thought that may help lighten the load a bit and I really don't like Yosh/Zyn - not their clunky, oversized and netbook-hostile GUIs anyway!

Did you try it with the ZynAddSubFX DSSI plug-in? Or do you need to change settings in real time?

Note that I was using my Focusrite 2i4 for mic input hence it would've been having to deal with a 32 (24) bit input - maybe I'll have to force 16-bit audio to vocode on something as puny as my netbook?

I don't think so, I think it should work with 32 (24) bit too.

Apart from the Sirlab vocoder I tried today, there is also the MDA and TAL vocoder plugins and apparently Rakarrack has a vocoder too. Has anyone tried all four? Which is the most CPU efficient? Which sounds the best? Which gives the best quality vs CPU usage? Any other vocoding tips you have to share?

I prefer the MDA Vocoder, it's better than the Sirlab one. Don't know about CPU usage, about the same as the Sirlab one I think.

[danboid]

I've never tried the zyn DSSI plugin actually. I'm presuming it has the same GUI as the standalone version of zyn and its the GUI that is the problem for the netbook and I.

I'll try again with the MDA plugin tonight and see if I have more luck with that. The sirlab ladspa plugin author admits in the readme he doesn't understand the code hence it could very likely be optimized and improved upon.

Have you tried the TAL and Rakarrack vocoders too Autostatic?

[autostatic]

I've never tried the zyn DSSI plugin actually. I'm presuming it has the same GUI as the standalone version of zyn and its the GUI that is the problem for the netbook and I.

It doesn't have a GUI at all. You can load presets and that's it.

Have you tried the TAL and Rakarrack vocoders too Autostatic?

Yes, but the TAL one crashed on me (64-bit plugin built with Juce framework + RT kernel == fail) and the Rakarrack one is not so good as the MDA one.

[danboid]

Thanks J!

I'll report back here with my results once I've tried the MDA vocoder.

[danboid]

Just had a quick go with the mda vocoder but no dice yet.

I installed mda-lv2 from the Deb Wheezy repos and it does include the vocoder plugin but no real, relevant docs.

I basically tried the steps outlined in your video but using mda instead of sirlab - would you expect that to work? Maybe I need to adjust a specific parameter or pick a different mda option before I get any output?

Thanks for your help autostatic!

[autostatic]

Hi Dan,

You might have to play around with the input ports, one is the formant and the other is the carrier. I'm not sure which one is which. I also vaguely recall that not a single synth was working in the MDA LV2 package and I'm not sure in how far the vocoder plug-in was affected by that (I also recall that this was fixed). It could also very well be that I'm using the LinuxVST version of the MDA vocoder. I need to check. Bear with me though, taking care of a baby and a toddler is pretty exhausting

[transmogrifox]

Another Vocoder I came across some time ago is the Snokoder. As far as vocoders go, it is really good, but a bit limited in that it synthesizes its own carrier, and has no option of inputting your own carrier signal. The input to Snokoder is MIDI.

As far as I know, nobody distributes a snokoder binary, so you need to compile source, but thanks to minimal external dependencies, Snokoder is easy to compile.

Here is the code, in the event you have trouble finding Snokoder source, I managed to find a copy online:
transformate.de/downloads/snokoder-v1.3-stable.c

I have thought (when I have time) to modify this source into a plugin and add the capability to have the option for an external carrier.

[danboid]

Sounds interesting - I'll have to try it but the MDA one sounds more like what I want as I want to route my own carrier too ideally.

Thanks for the tip Transmoggers!

[danboid]

Autostatic:

Have your kids left you alone for long enough to have a quick look at your vocoder setup yet? I'm going to have a more serious go at getting it working this weekend so it'd be a great help if you could highlight anything I may need to do differently from the way you routed the sirlab one in the above video and as you say I may need to install the VST version instead.

[danboid]

Autostatic said he had some vague memory of the MDA synths not working but I just tried the MDA ePiano and DX10 LV2 plugins under qtractor and they seemed to work fine.

Whilst doing that though I noticed I have a MDA Vocinput and a MDA vocoder LV2 plugin so now I'm wondering if I have (or it'd be preferable) to use the Vocinput plugin on the vocoder vocal input track?

[danboid]

I wrote to Mr LV2 himself yesterday - the one and only drobilla - and it was a great help as he clarified a few important things. First, the LV2 version of the MDA vocoder should function identically to the VST version hence the same docs apply, which are here:

http://mda.smartelectronix.com/vst/help/mdaplugs.htm

I've not verified it yet but the MDA plugins should have online help under A3 but this seems to be missing under qtractor, in which case you can just open /usr/lib/lv2/mda.lv2/Vocoder.ttl in a text editor and that contains the plugin docs.

Having read the docs and having all this explained to me I felt much more confident about trying the MDA vocoder and indeed I do have it working under qtr on my netbook using my Scarlett 2i4 mic and Noisemaker standalone as input (LD 3D Gameboy makes for a pretty good vocoder preset to use) now, the only problem I'm having is routing the audio from the vocoder track/bus into another track to record it - I've not had any luck with this yet.

Unlike the sirlab plugin, according to qtr the MDA vocoder outputs in stereo although I wouldn't have thought it'd really matter too much if I used/created a mono or stereo bus/track to route the vocoder to. I've tried all sorts of combinations of qtr bus and JACK IO routing here so far but not the right one to mix the vocoder to another qtr audio track. I'm presuming its me not getting the right path here rather than a bug in the plugin or qtr?

[danboid]

I've just tried with the mda VST vocoder plugin instead ( http://www.linux-vst.com/download/mda_linux.tar.gz ) but I'm having the same problem in that I can't seem to channel its output into a qtractor bus/track to record it. I'm trying to do it the way Autostatic does in his video ie

Create 2 stereo, duplex busses - called 'vocoder' and 'vocoder-mix' then assign the vocoder bus to the input and output of the first track, the one with the vocoder plugin. I attach my mic and synth to the inputs on the vocoder bus. Vocoder works if you monitor this track or route it to master.

Now I want to record the output of this to a new qtr track so I create a second track and assign 'vocoder-mix' as the input bus (leaving Master as the output) and I route the JACK audio outputs of the 'vocoder' bus to the inputs of this bus. Nothing shows up in the VU meter for this track after being connected/ armed / monitored. This happens with both the LV2 and VST versions of the mda vocoder under latest qtr svn running under Deb testing with the standard LV2 libs and using the mda-lv2 package from the repos.

[danboid]

I should've said that I can just hit record on the vocoder track and record its output but I get one vocoded channel and another 'raw' so ideally I would be able to just record the vocoded channel to a new (mono) track to spare editing and wasted space/time.

I've also had a very quick look at the TAL-vocoder but it didn't come with any presets like the LV2 version of MDA does and it looks like its got more of a learning curve. I don't think its quite as 'plug and play' as MDA and I failed to get any vocoded sound out of it in the 5m I've given it so far.

[danboid]

In Autostatic's video (see link in OP) when he is creating the buses he creates them both with 'Monitor (pass-through)' option enabled. It turns out that if you do that, the mda vocoder (LV2 or VST) doesn't work although this does kinda fix the issue I was having with no signal reaching my vocoder mix track. The signal does get routed when this option is enabled for the vocoder bus although the signal will arrive unprocessed/unvocoded.

So, as Rui recommends, it would seem I'd be better off running the mda vocoder under jalv or carla then recording its output to mhwaveedit or qtractor.