lsof |grep snd — how to free a Linux sound device 
Introduction
A program attempting to open a sound device for exclusive access may fail, if that audio device is already in use by another program. Also, depending on the capabilities of the audio device, such as whether it is full-duplex or not, errors may occur if capture is requested during playback. The actual error messages produced are program and operating system dependent. However, some typical error messages are listed below.
Audio device unavailable.
Cannot open audio device.
Could not gain access to /dev/dsp for writing.
Audio capabilities are not available on this machine.
Can't open output file '/dev/dsp': Device or resource busy.
Error Initializing Audio: There was an error initializing the audio i/o layer.
The reminder of this article briefly discusses the Linux sound architecture, followed by an explanation on how to locate and close applications that are using an audio device.
Linux audio
There are two main audio APIs used under Linux, namely the Open Sound System (OSS) and the Advanced Linux Sound Architecture (ALSA). There are also some alternatives, however OSS and ALSA are by far most popular audio interfaces. If you would like to read a bit more on how OSS and ALSA APIs compare then have a look at the ALSA vs. OSS article. In our article, we will mostly limit our discussion to the ALSA API. Note however that ALSA does provide a compatibility mode for OSS. Some very useful information about ALSA can be found in Alsa-sound-mini-HOWTO. Well, let me squeeze in another useful reference before proceeding. The beautiful thing about Linux is how simple things are under it. Under Linux, the audio devices are simply character devices. This makes it easy to access and use them. If you are planning on writing some code that requires audio device manipulation, then A Tutorial on Using the ALSA Audio API is well worth a read.
To start, lets have a look at audio cards installed on your system, by running the following command.
cat /proc/asound/cards
If, for instance, your sound card is based on NVidia NFORCE chip-set the output might look something like this:
0 [CK804 ]: NFORCE - NVidia CK804
NVidia CK804 with ALC850 at 0xd0103000, irq 233
On my system, the sound card is part of the Intel chip-set and is listed below as card 0. There is also another card with audio capabilities (card 1) which is an analog modem (in this case also a part of the Intel chip-set).
0 [I82801DBICH4 ]: ICH4 - Intel 82801DB-ICH4
Intel 82801DB-ICH4 with unknown codec at 0xe0100c00, irq 10
1 [Modem ]: ICH-MODEM - Intel 82801DB-ICH4 Modem
Intel 82801DB-ICH4 Modem at 0x2400, irq 10
Strictly speaking, an audio card could have a number of on-board devices. These devices are available under the /dev tree. The OSS device nodes are located in /dev/dsp* while ALSA uses /dev/snd/*. Lets have a closer look at ALSA's device nodes.
ls -la /dev/snd
The above command should produce an output similar to the one shown below, provided that you do have a sound card and that ALSA is present and loaded on your system.
total 0 drwxr-xr-x 2 root root 280 Oct 31 13:48 . drwxr-xr-x 12 root root 3840 Oct 31 13:49 .. crw------- 1 kamil root 116, 10 Oct 31 13:48 controlC0 crw------- 1 kamil root 116, 13 Oct 31 13:48 controlC1 crw------- 1 kamil root 116, 9 Oct 31 13:48 pcmC0D0c crw------- 1 kamil root 116, 8 Oct 31 13:48 pcmC0D0p crw------- 1 kamil root 116, 7 Oct 31 13:48 pcmC0D1c crw------- 1 kamil root 116, 6 Oct 31 13:48 pcmC0D2c crw------- 1 kamil root 116, 5 Oct 31 13:48 pcmC0D3c crw------- 1 kamil root 116, 4 Oct 31 13:48 pcmC0D4p crw------- 1 kamil root 116, 12 Oct 31 13:48 pcmC1D0c crw------- 1 kamil root 116, 11 Oct 31 13:48 pcmC1D0p crw------- 1 kamil root 116, 3 Oct 31 13:48 seq crw------- 1 kamil root 116, 2 Oct 31 13:48 timer
The above output shows a number of devices. Amongst others, there is the /dev/snd/controlC0 which is the control device for card 0, as well as /dev/snd/pcmC0D0p which is the raw audio playback device 0 on card 0. Note that the current naming convention is such that the card number follows the letter C and device number follows the letter D. The last letter on the pcm devices, c or p, indicates if the device is used for capture or playback, respectively.
Reproducing the error
Let us try to reproduce a conflict described in the introduction. First, we'll run mplayer to play an mp3 file song.mp3, as follows.
mplayer song.mp3
Then we'll try to play an audio file from Matlab.
>> [x, fs, nbits] = wavread('audio.wav');
>> sound(x, fs);
>> Unable to open the audio device for playback:
javax.sound.sampled.LineUnavailableException:
Audio Device Unavailable
The Java sound interface requires an exclusive access to the audio device, and hence the above error is produced. This is just one, somewhat artificial, example.
Freeing the audio device
So, the question is: how to free a Linux sound device? Well, the simplest way is to close, or end, the program that has opened it in the first place. A program will open an audio device with some read/write permissions and will hold a handle to that device. A well behaved application should release any open file handles (in-fact any resources) on exit back to the operating system. If an application fails to do that the operating system may collect these resources after the application terminates.
One may ask how to find out which program, or programs, have gained access to the audio device? That is where a little knowledge of Linux tools, and especially the lsof utility, comes in handy. Since we know where in the /dev tree the devices are mapped under ALSA we can use lsof, which stands for `list open files', to locate open sound device files. lsof will also tell us which programs have opened them.
So, lets run lsof and grep snd from its output to find out which sound devices are open.
lsof |grep snd
The output below shows that the KDE sound mixer, kmix, has opened the control devices for both cards (the sound card and the modem). More importantly, the mplayer is using the pcm device (i.e. the raw audio device) for playback. It is for this reason that the playback failed in Matlab.
kmix 5444 kamil 10u CHR 116,10 4807 /dev/snd/controlC0 kmix 5444 kamil 11u CHR 116,13 4944 /dev/snd/controlC1 mplayer 14004 kamil mem CHR 116,8 4684 /dev/snd/pcmC0D0p mplayer 14004 kamil 4r CHR 116,2 3914 /dev/snd/timer mplayer 14004 kamil 5u CHR 116,8 4684 /dev/snd/pcmC0D0p
Once the program that holds the audio device hostage has been identified, one can terminate it to free the device. If it is a shell program then ending it can be achieved with Ctrl+c. If it is a GUI based program then Alt+F4 may accomplish it, or f.e. File->Quit menu will also do it. Note however, that there are times when programs become unresponsive, or even crash, without properly freeing the open file handles. In such a case, the use of lsof becomes even more useful. Once the program is identified, a killall or kill commands can be used to terminate such a program. You could ask the application to close gracefully, i.e. give it time to perform any exit tasks (including properly freeing resources) by using
killall -TERM mplayer
or you could could force its termination by using
killall -SIGKILL mplayer
If you want to be more specific (and not kill every running instance of mplayer) then you could terminate a specific program using its process id (PID). Note that PID information is part of the lsof output. In our example the mplayer's PID is 14004. The command below was used to terminate it.
kill -TERM 14004
After terminating the program we can have another look at open audio devices.
lsof |grep snd
Notice, in the output below, that the raw audio device is no longer in use by mplayer.
kmix 5444 kamil 10u CHR 116,10 4807 /dev/snd/controlC0 kmix 5444 kamil 11u CHR 116,13 4944 /dev/snd/controlC1
Note that lsof is an extremely usesful utility. It finds many applications beyond the simple example described here. Another very useful utility is lslk — list local locks (i.e. list lock files). I will describe the use of lslk in a future article.
Did you find the above information useful and interesting? If so, please support this site by using the blog directory links at the bottom of this page. Thanks for your support!
If you have any Linux related problems or questions then please feel free to post them on our Linux Forums: http://linux.dsplabs.com.au/forums.
April 25th, 2008 at 1:03 am
when i run this command:
cat /proc/asound/cards
the output is:
no soundcars
But i belive that i have sound card coz my laptop sound is working when i use windows
help me
April 25th, 2008 at 3:20 am
what is the laptop model number and what is the sound card type? since your audio device is does not get detected by default it will be a fair bit trickier to get it to work… where I would start is googling to see if others have done it… for example I would search for "linux audio dell 9300 audigy sound blaster".
July 1st, 2008 at 10:11 pm
Excellent!… It worked!…
I had to do one step more: locating the program lsof using the command: whereis.
October 17th, 2008 at 5:40 pm
Thanks a bunch, I always resorted to closing down every program, or even rebooting. More uptime for me now!
October 28th, 2008 at 12:23 pm
wow! I've been searching the net to find solution to this problem. My transcription software sometimes runs and sometimes not. A hassle when transcription assignment arrives; but this article saved me! Thanks a lot and more power for giving this support!
December 4th, 2008 at 3:43 pm
solution in my pc
i did have to install libsnack to alsa
# apt-get install libsnack2-alsa
and solved.
tks.
January 20th, 2009 at 10:32 pm
I have no idea what Im doing here, I keep getting "no volume control gstreamer plugins and or devices found" as an error message when I click on my speaker icon!! Can anyone explain in LAYMEN terms how I fix this pleeeease????
January 21st, 2009 at 3:00 am
Kim, take a look at: http://linux.dsplabs.com.au/no-volume-control-gstreamer-plugins-and-or-devices-found-no-sound-or-voume-control-bug-on-ubuntu-p31/
April 23rd, 2009 at 5:10 pm
Indeed it worked!
This means that i can't hear music at the same time that
i want to speak with the microphone?
I had this problem for a long time, and after i used the commands above
for the first time my aMSN closed alone, and when i reopened it (without music)
the recording and the camera, worked properly. But when i re opened music
i had again the same error " Could not gain access to /dev/dsp for writing." .
Do i have to follow this procedure every time that i open the music?
PLz answer to my mail or here…
Thanks in advance
May 1st, 2009 at 9:36 am
First of all,this post is awsome and i've learned great and useful things but, i wonder if there's any way to be able to work with matlab (playing sounds) and listening music with another program (f.i. firefox and streaming) without having to kill one of both.
May 1st, 2009 at 11:44 am
Good question, lets try a few things… if I play music using mplayer and then use say play (a sox based playback script) to play another audio file. Both play fine.
If I play music using mplayer and the start matlab and try to play any audio (even after mplayer finished) it fails like so:
>> x=wavread('sp10.wav');
Warning: The playback thread did not start within one second.
> In audioplayer.playblocking at 35
In audiovideo/private/playsnd at 16
In sound at 46
In soundsc at 48
>> Unable to open the audio device for playback: javax.sound.sampled.LineUnavailableException: Audio Device Unavailable
To get it to work, I have to quit matlab and start it again. On the other hand, if I start matlab first and then try to play audio using mplayer, then mplayer fails to play — it simply sits there with the following prompt: Starting playback…
Ok, so as mentioned previously it is matlab's java based playback libs that simply requires exclusive access to the audio device. A bit of googling confirms that this is a common problem, and that java and/or alsa are at fault for not handling these things nicely.
Alright, if I had to do it, then my hack would be as follows. I would create a matlab wrapper function file called play.m somewhere in matlab's path. The content of play.m would be something like this:
% Saves audio data to a wav file before playback.
%
% audio - audio samples
% fs - sampling rate (Hz) [default: 8000]
% nbits - precision (bits) [default: 16]
% wavfile - filename for the wav file [default: '/tmp/tmp.wav']
%
% usage: play(audio, fs, nbits, wavfile)
function play(audio, fs, nbits, wavfile)
% parse input arguments or use defaults
switch(nargin)
case 1, wavfile='/tmp/tmp.wav'; nbits=16; fs=8000;
case 2, wavfile='/tmp/tmp.wav'; nbits=16;
case 3, wavfile='/tmp/tmp.wav';
case 4,
otherwise, error(sprintf('usage: play(audio, 8000, 16, "/tmp/tmp.wav"\ntype: help play for further details…'));
end
% save audio to file, scale prior to writing to avoid clipping
audio = 0.999*audio/max(abs(audio));
wavwrite(audio, fs, nbits, wavfile);
% playback the audio
if(exist(wavfile, 'file'))
syscmd = sprintf('mplayer %s', wavfile); % uses mplayer for playback
%syscmd = sprintf('play %s', wavfile); % uses sox play for playback
system(syscmd);
else
error(sprintf('%s: file not found.', wavfile));
end
% EOF
In matlab, help play gives the following:
>> help play
Saves audio data to a wav file before playback.
audio - audio samples
fs - sampling rate (Hz) [default: 8000]
nbits - precision (bits) [default: 16]
wavfile - filename for the wav file [default: '/tmp/tmp.wav']
usage: play(audio, fs, nbits, wavfile)
Overloaded methods:
audiorecorder/play
audioplayer/play
So make sure that your play.m is in the path before the other ones, or call it something else…
Ok, it is not pretty, but it works.
Kamil
May 8th, 2009 at 8:49 pm
Any solution for me?
May 9th, 2009 at 2:07 am
Hi Dimitris, in all likelihood your sound card can play and record at the same time, f.e. I can use, at the same time, mplayer for playback and gnome-sound-recorder for recording, without errors. The only thing that can get tricky are the mixer settings. Try un-muting all devices using alsamixer and set all the volumes to maximum to get started.
June 18th, 2009 at 2:28 pm
Kamil: completely agree with you, that is a so dirty solution, but as you point it works. The pitty is that Matlab/java and/or Alsa team doen't manage to fix it. Let's hope they get any solution. Linux is for work and have fun and multitasking system so, we should be able to do both, working with matlab while enjoying great music.
Anyway thanks for the tip, i liked itvery much, sorry for the absence but i've been working in other issues, and now i've already focused again in matlab : )
THNX again