sp - simple/sound pipe

Pretty good audio over UDP. Targeted for VM guest->host or guest->guest audio playback.

Can achieve better latencies than native Windows / passing through a soundcard.

This is new software, so please file an issue if you encounter any problems (compatability, etc.) :)

Features

• Low-latency
• Provides stable audio
• Provides 'accurate' audio
• Supports both sending and receiving on Windows and Linux
• Supports many backends (ALSA, PulseAudio, JACK (provided by libsoundio), native PipeWire, and WASAPI)
• Simple (4 source files for each built program and a header)

Contains 3 programs:

• sp-recv: Receives audio and plays it. For Linux and Windows. Has ALSA, PulseAudio, JACK (provided by libsoundio), native PipeWire, and WASAPI backends.
• sp-send: Captures audio and sends it to sp-recv. For Linux and Windows.
• sp-send-ll: An injected DLL that hooks the native WASAPI functions to yank the audio frames straight from the buffer of the hooked program before WASAPI gets it, therefore bypassing the whole Windows audio engine (mixer, resampler, etc.). This allows for 'perfect' latency. Windows-only (forever).

sp-send* sends the raw stream captured as soon as it gets it, without buffering, directly to sp-recv. sp-send-ll sends all streams separately, and has a separate instance for each application. This means that all separate streams will have their own separate stream in sp-recv at the receiver's end which enables (if your sound server supports it) routing and processing streams separately if you wish to do so.

Using

Grab the binaries in releases or follow the building steps in the 'Building' section below.

Start sp-recv <port> via a terminal on the machine or virtual machine that you want to receive audio to (the one that your physical sound system is connected to).

Start sp-send <receiver ip> <port> via a terminal on all of your machines or virtual machines that you want to send audio from.

sp-send on Linux requires PipeWire. After you start it, you will need to need to change your default output device to sp-send. Also, PipeWire's default config for VMs has bad latency. Make sure you edit your pipewire.conf and pipewire-pulse.conf (if you don't already have them, copy them from /usr/share/pipewire/ to ~/.config/pipewire/ and edit that) and set default.clock.min-quantum in vm.overrides to something like 128 and pulse.min.quantum to 128/48000 in each respective file. If you get crackling, you can try increasing the 128 to something higher or revert it back to 1024.

If you want to have low-latency audio and separated streams and are on Windows, run install-ll.bat as administrator (after reading 'Caveats' below). To uninstall, run uninstall-ll.bat as administrator. sp-send-ll requires both sp-recv and sp-send (non-ll) to be running and connected to work.

You can change the backends with '-o'. On Linux, I recommend the PipeWire or Jack backend. (Please don't use the PulseAudio it's shit. I get ~2 seconds of latency with it).

Building

Dependencies

For Linux:

• A C11 compiler supporting a small subset of GNU extensions (TCC works)

For sp-recv:

For ALSA, PulseAudio, and Jack backends:

• libsoundio
• speexdsp

AND/OR for PipeWire backend:

• PipeWire

For sp-send:

• PipeWire

For cross-compiling to Windows:

• MinGW
• meson
• CMake

For sp-recv:

• speexdsp

For sp-send:

• MinHook

speexdsp and MinHook are included as submodules for cross-compilation. For a native build, you want to install speexdsp and libsoundio, or pipewire via your system package manager.

Compilation for Windows is done by cross-compilation with meson, and CMake (via meson). Compiling for Windows on Windows is not supported, but shouldn't be difficult.

For the Linux build you do not need a build system. You can just run cc on recv.c and link against the libraries above.

For a quick build, there are some scripts included. Simply run ./build/linux && ./build/win. You will find the binaries located in build/.

If you are using ./build/linux, you can change the backends for sp-recv that are built by setting USE_SOUNDIO or USE_PIPEWIRE to 0 or 1. By default, only the PipeWire backend is built.

Caveats

sp is targeted towards VM communication, and is not optimized to be used over a physical network. The way that sp sends data is likely unfriendly to most routers and network cards. As mentioned earlier, sp does not do any extra buffering, batching, or processing at the sender's side. Therefore if you are using sp-send-ll, every hooked program will create it's own connection to send the stream. This means that with sp-send-ll, the more clients you have, the more bandwidth it uses.

sp-send-ll will not work with some protected programs (ex. web browser). It will also not work with audio capture programs (like OBS) as it mutes the stream to WASAPI when it gets hooked so you don't get double audio with sp-send.

Due to sp-send-ll's hooking nature, it requires being injected into every application. It therefore may be incompatible with some anticheat systems. The method of injection that is provided in releases is to use the AppInit_DLLs registry entries. These entries have also been used by malware as an attack vector. However, do note:

• Editing those entries requires administrator privileges
• The DLLs that are loaded will be placed in a protected folder that also requires administrator privileges to modify
• You should ideally be installing sp-send-ll only in a VM

I personally keep my sensitive information out of my VMs, so this is not a problem for me.

It may be possible to create a method that does not require AppInit_DLLs, however that will require finding the struct offset of the WAVEFORMATEX structure in IAudioClient (this is the way that Discord does its per-application audio sharing without being launched or hooked before the application runs). These offsets can differ with every Windows version, so will require a lot of maintainance. If someone knows a database of such offsets or an alternative method please let me know.

It is also currently possible to use a DLL injector to run a program with sp-send-ll . However this is inconvenient and requires the DLL to be injected and setup before the WASAPI audio client gets initialized.

(Note: this all is only with sp-send-ll. sp-send uses regular Windows APIs to grab the audio stream).

Todo

• Make the thing user friendly(-er)
• Optimize the header data

It was very simple at one point but now it's a less simple. Might tidy it up one day.

License

GPL v3.0