Question about Options/Preferences/Audio System

[dfkettle]

I don't know if this is expected behaviour or another issue with the Windows version of Hydrogen. When I go to Options/Preferences/Audio System, I don't see any audio devices listed (see screen capture). The sound is getting routed to my default audio device, a USB audio interface, but sometimes I want to send it to a virtual audio cable (I'm using EffectsRack for this). In other applications (Pure Data, Audacity, VCV Rack, etc.) I see the virtual audio cables listed under devices. But in Hydrogen, I don't see any devices (I have two real devices, the audio ports on the motherboard and an external USB interface, which is the default system device). Even if Hydrogen doesn't recognize or support the virtual audio cables, shouldn't I see at least the two real devices on my system and be able to choose which one I want to use?

If this is a bug in the Windows version, I'll open an issue on Github. Can somebody advise me on whether or not it's a bug?

[cme]

Yeah, there's currently no way to select a device -- noted here: #611.

I'm sure I remember there being some way to assign apps to sound devices in Windows as a workaround?

Selecting devices in general could probably be improved a lot, I think the only interface for it at the moment is specifying a textual name. We should refactor this so that the audio driver interfaces can provide a list of devices to the GUI to populate a menu with.

[cme]

Most audio software on Windows lets you choose which API to use. For example, Audacity (on my Windows 7 computer) lets me choose MME, Windows DirectSound or Windows WASAPI.

Aaah, this seems so obviously the right thing to do now it's been pointed out :) I was lulled into cul-de-sac by macOS having only a single host API that made me fail to consider it should just be an extra level of hierarchy.

But the error message doesn't really make any sense. My default device is a Steinberg UR22, which has 2 input and 2 output channels. And if I open Hydrogen without setting a device in Preferences (in other words, just let it default), this is the device that gets used, and it works fine! But if I try to select a device (any device, including the default device), I get the above error.

Great, thanks! I have a suspicion for what the issue is here. I note that "Line (3- Steinberg UR22)" doesn't explicitly say "line out"; is there also a line in that shows up with the same name? If so (and PortAudio is treating them as separate devices with the same name) then Hydrogen could mistakenly pick up the input device when trying to open the output stream.

Thanks again for your help here, this has been fantastically useful!

[dfkettle]

Great, thanks! I have a suspicion for what the issue is here. I note that "Line (3- Steinberg UR22)" doesn't explicitly say "line out"; is there also a line in that shows up with the same name? If so (and PortAudio is treating them as separate devices with the same name) then Hydrogen could mistakenly pick up the input device when trying to open the output stream.

There is a line in as well, but I'm not sure if would have the same name. Is it possible to filter out the input devices? That would probably reduce the duplication in the list, too. Some devices are showing up 3 times in the list for some reason. I don't think Hydrogen would mistake an input device for an output device, but the only way to be sure would be to look at the source code. It doesn't seem to matter which of the duplicate entries I select; they all seem to give an error.

[cme]

Would you mind giving this build a try, please?

https://ci.appveyor.com/api/buildjobs/2sjv13vkq91q9k09/artifacts/Hydrogen-1.1.0-win64.exe

(though note that it's a debug build so will pop open a console window to display the logging info)

[dfkettle]

I'm able to select the output device now, but the sound is very distorted for any output device except "Primary Sound Driver" or "Line (Steinberg UR22)". Note that my default device, a Steinberg UR22, is showing up 4 times in the list: 3 times as "Line (3 - Steinberg UR22" and once as "Line (Steinberg UR22". Other devices are also showing up 4 times, with one description appearing 3 times and a different description appearing once. I really think you need to find some way to filter out the duplicates or it will just confuse users. Sorting the list might be helpful too, as well as making it easier to remove duplicate entries. I've attached 3 screen prints showing the complete list of devices. (The third screen print overlaps with the second one.)

As for the distortion, it sounds as if there's a delay with feedback being applied, even though I'm not applying any effects at all. I've attached two sound files, 'test-good.txt' and 'test-bad.txt' (renamed because this forum only allows certain file types to be uploaded), to demonstrate. You'll have to rename them back to *.wav to listen to them (hopefully). I've also attached the pattern I used to test it (renamed as "Pattern.txt").
and the log file.

hydrogen.log
Pattern.txt
test-bad.txt
test-good.txt

[cme]

Other devices are also showing up 4 times, with one description appearing 3 times and a different description appearing once. I really think you need to find some way to filter out the duplicates or it will just confuse users.

Sorry, there's a small initial state bug there. The intention is that the device menu should now show only the device entries for the selected HostAPI. Without a HostAPI selected, it'll still show all of them. I have managed to verify this works as expected for me on a Windows VM (though slowly).

As for the distortion, it sounds as if there's a delay with feedback being applied, even though I'm not applying any effects at all.

Thaks for including the audio, that really helped. It looks like opening the non-default device isn't handling sample rate correctly, so I'll review that. Thanks again!

ETA: On closer inspection it's not the sample rate, but a timing or buffering issue that you're seeing, and it's consistent with one that I've managed to reproduce on a faster VM, and fix.

(The Pa_OpenDefaultStream call that's used (and was always the one used previously) when no device is specified, uses the device's 'high' default latency value, so that it's safe as a fallback. This is not documented in the API documentation. Using Pa_OpenStream and specifying a minimum latency, the latency is device-specific and may not be achievable, particularly for things like USB devices, in which case PortAudio will miss deadlines and emit silence for some buffer periods. The fix is to use the device default 'high' latency just like Pa_OpenDefaultStream)

If you feel like trying this, a release build with these changes is at:
https://ci.appveyor.com/api/buildjobs/deahbc1d2l5hou47/artifacts/Hydrogen-1.1.0-win64.exe

[dfkettle]

Hi! Sorry for the delay in responding, I didn't realize you'd created another build. It's great! The device list is a lot cleaner now, and the sound isn't distorted. I've only tested this on my Windows 7 computer, but I'll test it under Windows 8 & 10 as well, later today.

Thanks!

[dfkettle]

I tested it under Windows 8 & 10 and it works as well.

Would you happen to know the status of another issue I reported (#1147)? I'm not getting the error message I was before, but I'm not sure if it's actually been fixed. I have some plug-ins installed for other programs I'm using, but I'm not sure if any of them are LADSPA plug-ins (the only kind of plug-in supported by Hydrogen, is that right)?). If I can find some LADSPA plug-ins compiled for Windows, I can test that as well.

[theGreatWhiteShark]

BTW, some of the LADSPA effects were stereo, but most were mono. I tried stereo effects as well as mono. Should a mono effect be applied to both left and right channels, or will it just be ignored or cause an error?

The general/expected behavior would be that the mono effect is opened twice by the application for both the left and right channel and the program just shows a single window to control the parameters of both of them. But I wouldn't be surprise if Hydrogen handles it not as expected.

Personally, I think the plugins support in the mixer is close to embarrassing. One can only use aux sends instead of adding effects to single tracks. While having aux sends is in general a good idea, the total number of 4 is extremely limiting. (I need an EQ for each instrument to cut out all the resonance and clean up improper recording before I can use a kit for more than a rough demo). The overall UI is super odd. Why faders for the parameters with a meter that is not moving at all but just represents the current value? It's not adding any information but might be misleading. In the long run we might rework the whole thing but it's not on the priority list right now.

[cme]

After extensive googling, I found one post stating that you had to define an environment variable named LADSPA_PATH and set it to the directory containing the plugins (I don't think this is mentioned in the official Hydrogen manual).

We really ought to add some sort of GUI for this. Dependencies on environment variables are bad especially on Windows and macOS where users aren't even expected to know what those are.

[dfkettle]

We really ought to add some sort of GUI for this. Dependencies on environment variables are bad especially on Windows and macOS where users aren't even expected to know what those are.

@theGreatWhiteShark @cme Agreed. But in the meantime, here are instructions for Windows 7, 8 and 10. BTW, Audacity is still 32-bit, so any plugins included with it will be 32-bit. So they won't work in Hydrogen (unless you downloaded the 32-bit version of Hydrogen, I guess). That's why I had to download Radium. Would it be possible to bundle the effects (since they're all free, although they may not all be open-source) and include them in a sub-directory of Hydrogen (say <Hydrogen install directory>\LADSPA)? Then you wouldn't have to set any environment variables, since Hydrogen would just assume they're in that directory.

Windows 7:

1. Click on "Start" (button in lower left corner).
2. Click on "Control Panel".
3. Click on "System and Security".
4. Click on "System".
5. Click on "Advanced system settings".
6. Click on "Environment Variables".
7. Under either "User variables" (for yourself only) or "System variables"
   (for all users), click on "New...".
8. For "Variable name", enter "LADSPA_PATH".
9. For "Variable value", enter the full path to directory containing plugins.
10. In "Environemt variables" window, click on "OK".
11. In "Advanced system settings" window, click on "OK".
12. In "System" window, click on "OK".
13. Close "System and Security" window.
14. If Hydrogen is open, close and re-open.

Windows 8:

1. Right-click on the "Start" (button in lower left corner).
2. Go to step 2 above.

Windows 10:

1. Right-click on the "Start" (button in lower left corner).
2. Click on "Settings".
3. Go to step 5 above.

[dfkettle]

Correction: Seems Audacity is 64-bit, so I'm not sure why the plugins included with it weren't recognized. Anyway, it seems the same plugins are included with Radium, and they were recognized.

[theGreatWhiteShark]

Also the LMMS doc suggests it is bundled and shipped with the most essential LADSPA plugins.

[cme]

Funny thing, I've never actually tried using LADSPA effects, but... yeah, I don't hear any difference. A quick look at the code and... nothing ever seems to actually fill the input buffers of the LADSPA filters. They're connected up in audioEngine_setupLadspaFX with the same buffers as input and output, which doesn't seem right either, but it's been this way at least since 2014. I'll have a dig into the history to try and make sense of this. @theGreatWhiteShark you've used LADSPA effects, right?

[cme]

Well, I could easily be misunderstanding how it works, but it looks to me like it is broken* and has been so for a long time.

Any synthesis type plugin that creates an output but doesn't require an input (are there any of these?) would work, because the output of the plugin is added into the audio output. But this seems wrong, doesn't it? I thought all plugins were supposed to function as filters.

I wouldn't be surprised if I'm just entirely misunderstanding it.

[dfkettle]

In the VST world, VSTi's (short for Vst instruments) don't require any audio input (they receive MIDI data instead). I don't know about LADSPA, but these are supposed to be effects, so I think, by definition, they require input.

[theGreatWhiteShark]

Any synthesis type plugin that creates an output but doesn't require an input (are there any of these?) would work, because the output of the plugin is added into the audio output. But this seems wrong, doesn't it? I thought all plugins were supposed to function as filters.

There are a number of plugins that are agnostic to their input, like https://tomscii.sig7.se/tap-plugins/ladspa/pinknoise.html . But I think the general concept is that they receive an input buffer, do some processing, and hand it over to the next element in the processing chain. In principle it should be possible to do some wave synthesize on the basis of an empty track with e.g. http://quitte.de/dsp/caps.html#Sin . But nowadays you would use a more convenient graphical LV2 plugin.

@theGreatWhiteShark you've used LADSPA effects, right?

I use/used them a lot (but mostly outside of Hydrogen). I just tried at 4355e93 and it works fine.

[cme]

So, it turns out that I just forgot how these are wired up (in both code terms, and in GUI terms).

In the code, the filter input buffer is written by the sampler on a per-note basis, before the mixer.

Correspondingly, in the GUI, each instrument's send level for each filter has to be set (defaulting to 0). The GUI isn't very helpful and I don't think it's clear in the manual. But once I remembered that, it all seems to work fine.

Video, Hydrogen on Windows with a LADSPA filter

The latency is horrible but that's partially due to running in a VM, and partially due to #1352 . The low audio qualiy and background noise is due it being acoustic from my laptop's speakers to microphone.

[dfkettle]

Correspondingly, in the GUI, each instrument's send level for each filter has to be set (defaulting to 0). The GUI isn't very helpful and I don't think it's clear in the manual. But once I remembered that, it all seems to work fine.

OK, now I understand. You have to turn up one of the four tiny knobs as well as the knob labelled "RETURN". I'm not sure if there's a latency problem, but I hear the effect now.