If you want to have different programs sending their output to the same sound card (like the HiFiBerry DAC+) on the Raspberry Pi you might run into problem. The I2S sound system allows only exclusive access to the sound subsystem. The solution for this and some other issues is PulseAudio. It creates an additional abstraction layer that allows different sound sources to use is simultaneously.
unfortunately the Raspbian pulseaudio package does not include some resamplers. I was not able to use the “ffmpeg” resampler with it. Therefore I decided to build my own pulseaudio package. This also allows to use PulseAudio 5 instead of version 2 that it part of the Raspbian distribution.
Note that you need to install a lot of prerequisites before you start to configure and compile pulseaudio:
sudo apt-get install -y libltdl-dev libsamplerate0-dev libsndfile1-dev libglib2.0-dev libasound2-dev libavahi-client-dev libspeexdsp-dev liborc-0.4-dev libbluetooth-dev intltool libtdb-dev libssl-dev libudev-dev libjson0-dev bluez-firmware bluez-utils libbluetooth-dev bluez-alsa libsbc-dev libcap-dev
This will install not only these packages, but also a lot of dependencies. Note that I did not install the X11 headers as Pulseaudio will run in systems mode.
Now you can get, extract, configure pulseaudio.
tar xvf pulseaudio-5.0.tar
Now check the output of teh configuration process. It will look like this:
System Runtime Path: /usr/local/var/run/pulse
System State Path: /usr/local/var/lib/pulse
System Config Path: /usr/local/var/lib/pulse
Compiler: gcc -std=gnu99
CFLAGS: -g -O2 -Wall -W -Wextra -pipe -Wno-long-long -Wno-overlength-strings -Wunsafe-loop-optimizations -Wundef -Wformat=2 -Wlogical-op -Wsign-compare -Wformat-security -Wmissing-include-dirs -Wformat-nonliteral -Wpointer-arith -Winit-self -Wdeclaration-after-statement -Wfloat-equal -Wmissing-prototypes -Wredundant-decls -Wmissing-declarations -Wmissing-noreturn -Wshadow -Wendif-labels -Wcast-align -Wstrict-aliasing -Wwrite-strings -Wno-unused-parameter -ffast-math -fno-common -fdiagnostics-show-option
CPPFLAGS: -DFASTPATH -D_FORTIFY_SOURCE=2
LIBS: -lcap -lrt -ldl -lm
Enable X11: no
Enable OSS Output: yes
Enable OSS Wrapper: yes
Enable EsounD: yes
Enable Alsa: yes
Enable CoreAudio: no
Enable Solaris: no
Enable WaveOut: no
Enable GLib 2.0: yes
Enable Gtk+ 3.0: no
Enable GConf: no
Enable Avahi: yes
Enable Jack: no
Enable Async DNS: no
Enable LIRC: no
Enable Xen PV driver: no
Enable D-Bus: yes
Enable BlueZ 4: yes
Enable BlueZ 5: yes
Enable udev: yes
Enable HAL->udev compat: yes
Enable systemd login: no
Enable systemd journal: no
Enable TCP Wrappers: no
Enable libsamplerate: yes
Enable IPv6: yes
Enable OpenSSL (for Airtunes): yes
Enable fftw: no
Enable orc: yes
Enable Adrian echo canceller: yes
Enable speex (resampler, AEC): yes
Enable WebRTC echo canceller: no
Enable gcov coverage: no
Enable unit tests: no
simple database: no
System User: pulse
System Group: pulse
Access Group: pulse-access
Enable per-user EsounD socket: yes
Force preopen: no
Preopened modules: all
Legacy Database Entry Support: yes
Make sure you have everything that you need compiled in. The alsa-module is the most important, but you will also need at least libsamplerate.
Now compile and install pulseaudio.
sudo make install
It will be installed to /usr/local. If you compile it at the Raspberry Pi, you have some time for a coffee. Crosscompiling on a PC is much faster, but also much more complicated, therefore I would not recommend this.
I would recommend to start with a minimal set of modules:
load-module module-native-protocol-unix auth-anonymous=1
Now start Pulseaudio:
/usr/local/bin/pulseaudio -vvv --system --disallow-exit --disallow-module-loading=1 --high-priority
Pulseaudio will run in foreground. This has to be changed later, but it makes debugging much easier.
I usually use mplayer with some test .wav and .flac files like this:
pi@raspberrypi ~ $ mplayer ./test-44.wav
MPlayer svn r34540 (Debian), built with gcc-4.6 (C) 2000-2012 MPlayer Team
mplayer: could not connect to socket
mplayer: No such file or directory
Failed to open LIRC support. You will not be able to use your remote control.
libavformat version 53.21.1 (external)
Mismatching header version 53.19.0
Audio only file format detected.
Load subtitles in ./
Opening audio decoder: [pcm] Uncompressed PCM audio decoder
AUDIO: 44100 Hz, 2 ch, s16le, 1411.2 kbit/100.00% (ratio: 176400->176400)
Selected audio codec: [pcm] afm: pcm (Uncompressed PCM)
AO: [pulse] 44100Hz 2ch s16le (2 bytes per sample)
Video: no video
A: 35.0 (34.9) of 60.0 (01:00.0) 3.9%
Have a look at the output. If Pulseaudio is working correctly, you will see the AO: [pulse] as output device.
However there are some issues, the major one is resampling. To make sure that different sound sources can be mixed all have to be synchronized to a common sample rate. This is not a big problem if you use only 44.1 and 48kHz, but if you also want to playback high-resolution music you will run into trouble with the Raspberry Pi. What is the reason for this? The biggest issue is that the Raspberry Pi CPU is not very powerful and resampling is a relatively complex procedure.
If you read about this issue on several pages on the Internet, many people suggest to use the “trivial” resampling method. It does not need a lot of calculations and therefore works on the Raspberry Pi without a lot of problems. However how trivial is trivial? Have a look at this sine wave:
My audio analyzer shows almost 3% distortions! This is not a good method to use if you’re interested in high-quality audio.
Ok, let’s look at other resamplers. The “src-sinc-…” resamplers are too complex. Even the “src-sinc-fastest” is too complex for 192kHz material.
The “src-zero-order-hold” resampler performs as bad as “trivial”.
The two resamplers that perform quite well are “src-linear” and “ffmpeg”. While distortions are a bit less with “src-linear”, “ffmpeg” used less CPU. As distortions are still quite low with “ffmpeg” I recommend this for resampling.
Upsampling or downsampling?
Ok, we found a resampler that works. But what should be the target sample rate? Down to 44.1 or 48kHz or up to 192kHz? Some people will argue that frequencies above 20kHz are not audible and therefore 44.1kHz or 48kHz are enough. While I don’t want to discuss pros and cons of high-resolution audio format here, there are some other details to think about.
With a sound card like the HiFiBerry DAC+, the clocks are generated from the Raspberry Pi. This works quite well, but different sample rates provide better clocks to the DAC than others. The DAC shows the best performance at higher sample rates. Therefore the distortions will be lower with upsampling to 192kHz than with downsampling to 44.1kHz. There is also another effect: Playback of lower sample rates does not need a lot of processing. There is more processing possible than with 192kHz material. Playing back 192kHz material with its native sample rate does not need resampling. This balances processing power between the playback software and Pulseaudio.
My daemon.conf configuration file now looks like this:
resample-method = ffmpeg
enable-remixing = no
enable-lfe-remixing = no
default-sample-format = s32le
default-sample-rate = 192000
alternate-sample-rate = 176000
default-sample-channels = 2
All other settings use the defaults.
Setting up Pulseaudio on the Raspberry Pi with support for high-resolution sound formats is not trivial. However by using the right sample rates and the right resampler, it will perform well even on the Raspberry Pi. With the configuration shown here Pulseaudio uses about 50% CPU when playing back 44.1, 48kHz and 96kHz material and 20% during playback of 192kHz streams.