Sound Pusher — real-time encoded audio output for OS X

Provides a virtual 5.1-channel audio output device whose contents are real-time encoded to AC3 and sent as digital SPDIF stream to a real audio device output.

WARNING: Please turn down the speaker / headphone volume before running this application, as any component misinterpreting a compressed audio stream as "normal" audio signals may result in unexpectedly loud noise.

Requirements

The application requires macOS 14.2 "Sonoma" due to use of Core Audio taps to capture the system audio. It requires a compatible audio device capable of outputting compressed AC3 audio formats (format IDs ac-3, cac3, IAC3, iac3). All testing was done with cac3 (aka kAudioFormat60958AC3).

Usage

There are two components required to make this work:

1. SoundPusherAudio — A user-space (sand-boxed) CoreAudio driver (aka AudioServerPlugin) that provides a 5.1 channel output stream whose contents can then be captured. SoundPusherAudio.driver must be installed (copied) into the /Library/Audio/Plug-Ins/HAL directory. This driver was previously called Loopback Audio and you may have to remove the old version (called LoopbackAudio.driver) manually. Note that if this device provides audio inputs, you may have to allow microphone access in addition to system audio recording permission. Older versions of this driver used to mirror the output it receives back to its (virtual) microphone input. The last release that included this loopback functionality was v1.4.2. Feel free to use that version of the driver, but I'd recommend you look at Rogue Amoeba's Loopback since I'm not interested in supporting that use case.
2. SoundPusher — An application that continuously reads audio from SoundPusher Audio (or any other suitable) device, encodes it into a compressed format, and then sends that compressed stream to a (hopefully) real sound device's digital output stream. This is controlled through its menu bar extra.

Once the driver is installed and SoundPusher running, you should be able to select the desired combination of supported source and digital output devices (which will forward any sound output from the source device to the designated digital output). SoundPusher will set the default output device to the selected source device (e.g. SoundPusher Audio) and restore the default device on exit.

You can override the default IOCycle timing adjustment (which attempts to postpone the encoding to as late as possible to just before the hardware needs the next packet) in case you notice sound drop-outs or overloads. The default is a safety-factor of 8. This is done via defaults write de.maven.SoundPusher IOCycleSafetyFactor -float <value>. Any value less than 1 disables IOCycle adjustment altogether, which means that CoreAudio starts preparing the next packet roughly when the current packet is sent to the hardware (which is usually a bit of a waste as current hardware can easily encode 100x faster than real-time).

You can also switch back to the old rear channel upmixing with defaults write de.maven.SoundPusher UpmixDPLiiRear -bool NO.

Contact & Support

Please report any issues on GitHub.

Dependencies

FFmpeg — libavcodec, libavformat, libswresample

Here's a script that builds a cut-down version of FFmpeg that includes all that's required for SoundPusher:

#!/bin/sh
MACOSX_DEPLOYMENT_TARGET=10.14 export MACOSX_DEPLOYMENT_TARGET
# when compiling for x86_64 on Apple Silicon (aarch64) you probably want to install yasm and add the following:
# --extra-cflags="-target x86_64-apple-macos10.14" --extra-ldflags="-target x86_64-apple-macos10.14" --arch=x86 --x86asmexe=<PATH_TO_YASM>/bin/yasm
./configure --prefix=$FFMPEG_HOME --cc=clang --extra-cflags="-fno-stack-check" --disable-static --enable-shared --disable-all --disable-autodetect --disable-programs --disable-doc --disable-everything --disable-pthreads --disable-network --disable-dct --disable-dwt --disable-lsp --disable-lzo --disable-rdft --disable-faan --disable-pixelutils --enable-avutil --enable-avcodec --enable-avformat --enable-swresample --enable-encoder=ac3 --enable-muxer=spdif
make -j8
make install
# update shared library install names
cd $FFMPEG_HOME/lib
AVUTIL_ID=`otool -D libavutil.dylib | tail -1`
AVCODEC_ID=`otool -D libavcodec.dylib | tail -1`
AVFORMAT_ID=`otool -D libavformat.dylib | tail -1`
SWRESAMPLE_ID=`otool -D libswresample.dylib | tail -1`
for lib in libavutil libavcodec libavformat libswresample
do
    install_name_tool -id @rpath/$lib.dylib -change $AVUTIL_ID @rpath/libavutil.dylib -change $AVCODEC_ID @rpath/libavcodec.dylib -change $AVFORMAT_ID @rpath/libavformat.dylib -change $SWRESAMPLE_ID @rpath/libswresample.dylib $lib.dylib
done

If you're building a universal binary (for Intel (x86_64) and Apple Silicon (aarch64)) you probably want to compile FFmpeg twice and then lipo -create -output <outLib> <inLibA> <inLibB> all the resulting libs together. Note that you should add FFMPEG_HOME in Xcode's Preferences -> Locations -> Custom Paths and point it towards the install directory of FFmpeg so that the Xcode project can find it.

Acknowledgements

This software is built on the experience of others:

• SoundPusherAudio (formerly known as LoopbackAudio) is based on the NullAudio user-space driver provided by Apple as example code.
• TPCircularBuffer by Michael Tyson is used to move audio and packet data around.
• FFmpeg is used to encode to compressed formats and then mux those packets into an SPDIF compliant bit-stream.
• Joseph Lyncheski's Core Audio Tap example