Add WASAPI loopback implementation

[WeekendWarrior1]

Re implements this patch: audacity/audacity@be7467c#diff-6ea9c8ba5ecef609fc4cfa9fbbb1bc11ad66acb997565df06ad6ce1562a10144 to add loopback for WASAPI devices.

I'm very new to C so I welcome any feedback or improvements to this PR.

Closes this:
#668

And would really help python-sounddevice users:
spatialaudio/portaudio-binaries#6
spatialaudio/python-sounddevice#287
tlecomte/friture#64
spatialaudio/python-sounddevice#281
spatialaudio/python-sounddevice#14

Also see this project which maintains a fork of portaudio with loopback support specifically for PyAudio: https://github.com/intxcc/pyaudio_portaudio

[RossBencina]

@dmitrykos could you review this please

[WeekendWarrior1]

This was marked as a draft because shauneccles discovered some issues when using other devices after testing this patch, will bug hunt tonight.

[dmitrykos]

@RossBencina yes, I will do so once have a bit of free time, I am tracking this PR already since it was proposed. I also will wait for announced @WeekendWarrior1's improvements.

[RossBencina]

Awaiting response to Dmitry's change requests.

[dmitrykos]

@not-matt thank you for the changes.

The implementation can be improved further by calculating the number of available eRender devices in order to allocate memory once here:

portaudio/src/hostapi/wasapi/pa_win_wasapi.c

Line 2294 in 2a68aeb

if ((paWasapi->deviceCount != 0) && ((deviceInfoArray = AllocateDeviceListMemory(paWasapi)) == NULL))

Otherwise memory is wasted (not leaked completely but it is inefficient anyway) when you reassign it here:

portaudio/src/hostapi/wasapi/pa_win_wasapi.c

Line 2343 in 2a68aeb

if ((hostApi->deviceInfos = (PaDeviceInfo **)PaUtil_GroupAllocateMemory(paWasapi->allocations,

If memory is allocated once by AllocateDeviceListMemory then AddLoopbacksToDeviceList could be minimized to only fill the loopback-specific info inside the

portaudio/src/hostapi/wasapi/pa_win_wasapi.c

Line 2301 in 2a68aeb

for (i = 0; i < paWasapi->deviceCount; ++i)

loop.

It would make implementation much simpler.

Here is helper function which will return the number of eRender devices in the device list:

// ------------------------------------------------------------------------------------------
static UINT32 GetRenderDeviceCount(IMMDeviceCollection *pEndPoints)
{
    HRESULT hr;
    UINT32 deviceCount;
    IMMDevice *device;
    IMMEndpoint *endpoint;
    EDataFlow flow;
    UINT32 ret = 0;

    hr = IMMDeviceCollection_GetCount(pEndPoints, &deviceCount);
    IF_FAILED_JUMP(hr, error);

    for (UINT32 i = 0; i < deviceCount; ++i)
    {
        hr = IMMDeviceCollection_Item((IMMDeviceCollection *)pEndPoints, i, &device);
        IF_FAILED_JUMP(hr, error);

        if (SUCCEEDED(hr = IMMDevice_QueryInterface(device, &pa_IID_IMMEndpoint, (void **)&endpoint)))
        {
            if (SUCCEEDED(hr = IMMEndpoint_GetDataFlow(endpoint, &flow)))
                ret += (flow == eRender);

            SAFE_RELEASE(endpoint);
        }

        SAFE_RELEASE(device);
    }

    return ret;

error:

    return 0;
}

[not-matt]

Hi @dmitrykos, thank you for the feedback. I've worked with @WeekendWarrior1 in the past but I've not been able to reach him to wrap up this PR.

I can make an attempt to implement your feedback, but I have no experience writing C. If you have the time, would you consider writing these changes yourself?

I'm in touch with @shauneccles and can get any changes you would make added to this PR very quickly.

[dmitrykos]

@not-matt ok, I will try to implement in the way I advised and let's see how it works then.

[not-matt]

Nice, cheers. Do let me know if there's anything I can help you with, especially testing.

[nidefawl]

I have been waiting for this feature to land upstream. I maintained my own version for a couple of years.
I will help testing once @dmitrykos is done with the implementation

[dmitrykos]

I made necessary cleanup changes, would you please test how it works on your side. I tested Win32 and UWP modes and it works quite well, no problems noted.

Loopback mode does not support Exclusive mode though that is quite natural because it is a proxy for a playback device.

If your test results confirm that all runs fine I think we could finally merge it.

[not-matt]

Good stuff! I'm away this weekend, I'll take a look once I'm back.

[nidefawl]

I built and linked against this PR and the loopback input works nicely!

I haven't tested the PaWasapi_IsLoopback function yet

[nidefawl]

PaWasapi_IsLoopback works as intended

[dmitrykos]

@nidefawl great, thank you for testing!

[shauneccles]

Can confirm it behaves as expected over here on Win 10/Win 11.

[dmitrykos]

@shauneccles thank you for testing. I added comments into the source code regarding [Loopback] loopback device name identificator usage. If the form of this identificator is unexpected by some of your projects, please comment. I assumed that Windows audio device names often have device description in (), so to differentiate identificator more I used [Loopback]. It can be adjusted of course if such change created some problems.

[shauneccles]

No issues with using square brackets that I can see/think of, it makes sense to keep them different to give people another signal that it is a loopback device.

[not-matt]

+1, this works well for me on Windows. Haven't ran into any issues.

[dmitrykos]

Thank you for proposing the PR and tests, I think it can go into the main repo!

Closes #668

[Be-ing]

Thanks for taking the time to do this right and get it upstreamed!

[bear101]

After adding loopback devices to WASAPI it looks like PaHostApiInfo::defaultInputDevice and PaHostApiInfo::defaultOutputDevice are broken. I haven't quite figured out why the indexing is not matching.

#include <portaudio.h>
#include <iostream>

int main(int argc, char* args[])
{
    PaError err = Pa_Initialize();

    int n_devices = Pa_GetDeviceCount();

    for (int i=0;i<n_devices;++i)
    {
        const PaDeviceInfo* devinfo = Pa_GetDeviceInfo(i);
        std::cout << "#" << i << " " << devinfo->name << std::endl;
    }

    PaHostApiIndex hostapiIndex = Pa_HostApiTypeIdToHostApiIndex(paWASAPI);
    const PaHostApiInfo* hostapi = Pa_GetHostApiInfo(hostapiIndex);

    std::cout << "Default WASPI input #" << hostapi->defaultInputDevice << std::endl;
    std::cout << "Default WASPI output #" << hostapi->defaultOutputDevice << std::endl;
    return 0;
}

Output:

#0 Microsoft Sound Mapper - Input
#1 Microphone (High Definition Aud
#2 Microphone (HD Pro Webcam C920)
#3 Microsoft Sound Mapper - Output
#4 Headphones (High Definition Aud
#5 VE248 (Intel(R) Display Audio)
#6 Primary Sound Capture Driver
#7 Microphone (High Definition Audio Device)
#8 Microphone (HD Pro Webcam C920)
#9 Primary Sound Driver
#10 Headphones (High Definition Audio Device)
#11 VE248 (Intel(R) Display Audio)
#12 VE248 (Intel(R) Display Audio)
#13 Headphones (High Definition Audio Device)
#14 Microphone (High Definition Audio Device)
#15 Microphone (HD Pro Webcam C920)
#16 VE248 (Intel(R) Display Audio) [Loopback]
#17 Headphones (High Definition Audio Device) [Loopback]
Default WASPI input #16
Default WASPI output #14

Something must be going wrong with the indexing either here and here.

[dmitrykos]

@bear101 thank you for reporting, I will have a look shortly.

[dmitrykos]

@bear101 would you please check #688 PR regarding the problem you noticed.

[GiuseppeUFSM]

Hello Guys, is there any exemple of the loopback feature implementation?
I'm asking bc I didn't find it and, as I'm learning now how to use PortAudio, it would be extremely helpful!
Thanks!

[dechamps]

@GiuseppeUFSM You don't need any special code to use this feature. Just use PortAudio normally and the "virtual" loopback devices will just naturally show up in the usual list of input devices. In fact it should even work in existing applications if you replace their PortAudio DLL.