PulseAudio 3.0 Released With New Sound Features

[ShadowBane]

the sound quality difference should be unnoticeable, if not the same (probably is the same)

im just saying for the people who think its any different, that its probably not
(unless the settings are bad, either in PA or the program using PA/alsa)

PS i used ladspa, not dmix

Ladspa is a plugin format... If you were doing any software mixing with alsa it was using dmix.

[drag]

Also it's worth pointing out that hardware mixing is still hardware mixing. And don't fool yourself for a second into thinking that the hardware is more sophisticated then the software.

It terms of audio quality from roughly from best to worst:

1. 2-channel PCM or multichannel HDMI passthrough
2. decoded audio out, no mixing. (Pulseaudio decides on the fly if mixing is necessary)
3. remixed digital/analog out using Pulseaudio
5. remixed digital/analog out using audio card hardware
6. remixed digital/analog out using Alsa dmix.

[drag]

dont compare mp3 music, to flash video from youtube to see CPU usage man...
you have something broken i can play 10 streams of audio with PA, and my usage is 1% no more

If you are using Alsa-only and have the ability to play multiple streams on common modern hardware then you are using dmix plugin for software mixing. The software mixing for Alsa-only isn't going to show up the same using things like 'top' when you are trying to compare software mixing for PulseAudio. The cpu usage is still there, but there will be less evidence of it.

On top of that the software mixing that PulseAudio does is much higher quality then the software mixing 'dmix' plugin does. The audio output is much higher quality. If you can't tell the difference and want to reduce the cpu usage you can choose a different mixing algorithm.

I believe that resampling-method trivial is the same as what Alsa by itself uses, which is the worst quality one that PA supports.

[NomadDemon]

"In a typical installation scenario under Linux, the user configures ALSA to use a virtual device provided by PulseAudio. Thus, applications using ALSA will output sound to PulseAudio, which then uses ALSA itself to access the real sound card."

thats one copy that dosent occur with alsa, wouldnt occur at all if pulseaudio was kernel-level

hmmm just been reading about how it was desinged
in theory it should be good, and with mr. Poettering orchestrating it it can achieve what it says

but when i see pulseaudio using 5% of my laptops cpu(could be just ubuntu ppl screwing it up again) to play a low sample rate sound from a youtube video, that tells me its either not there yet or its just doing a lot of useless stuff (5% of a cpu is a whole lot, lot more then people today think)

zero-copy is also a kernel thing that glibc can use, but still the page says that PA has a scheduler running all the time
all in all it can never achieve the efficiency of alsa (alsa with a ladspa plugin for a equaliser used ~1.5cpu to play a 44k mp3 on that shitty laptop)


and for all you talking about sound quality, please dont

dont compare mp3 music, to flash video from youtube to see CPU usage man...
you have something broken i can play 10 streams of audio with PA, and my usage is 1% no more

[benmoran]

Since we're saying random things about Pulse that have nothing to do with 3.0 I'll throw in my 2c.

On my media center, Pulse was a horrible experience that I ended up pulling off. The only way to make it work with my hdmi sound output was to replace the udev detection with loading a module specifically for 0,7 or whatever the proper output was. (HDMI would show up, but it would be trying to use 0,3 or something and not work at all). On top of that xbmc crashed frequently as the sound device would disappear - removed pulse and those crashes went away, and sound was consistent.

Those problems are almost definitely caused by XBMC itself, and not PulseAudio. The newest releases of XBMC work flawlessly in this regard. I use it myself. In addition to HDMI audio now working without hacking it, XBMC also now supports a few new surround formats by default.

[GreatEmerald]

the sound quality difference should be unnoticeable, if not the same (probably is the same)

If you set PA to use trivial resamping (the fastest method), you can hear the difference immediately. The sound gets "fuzzy". Very noticeable when KDE default sounds are being played.

[TheCycoONE]

Since we're saying random things about Pulse that have nothing to do with 3.0 I'll throw in my 2c.

On my media center, Pulse was a horrible experience that I ended up pulling off. The only way to make it work with my hdmi sound output was to replace the udev detection with loading a module specifically for 0,7 or whatever the proper output was. (HDMI would show up, but it would be trying to use 0,3 or something and not work at all). On top of that xbmc crashed frequently as the sound device would disappear - removed pulse and those crashes went away, and sound was consistent.

On my laptop on the other hand I use a USB headset and an external sound card depending on what I'm doing. Pulse makes all that incredibly seamless - far nicer experience than editing asoundrc files and using alsactl every time I want to switch.

As a result I have mixed feelings - I can see how it could cause very polarizing viewpoints depending on a person's hardware and common use cases. I'm curious to try it again on the media center with xbmc 12 and PulseAudio 3 and see if they've fixed up the rough edges.

[gens]

Is there something wrong with Pulseaudio's sound quality? The main reason that it uses more CPU time than dmix does for the same job is that uses a better (and more computationally expensive) re-sampler when re-sampling is needed. Another thing that could cause extra cpu usage is doing more re-sampling than needed (setting alternate sample rates in pulse can help with this, if only one thing is running it will use a sample rate that is less expensive to use instead of the primary one.)

the sound quality difference should be unnoticeable, if not the same (probably is the same)

im just saying for the people who think its any different, that its probably not
(unless the settings are bad, either in PA or the program using PA/alsa)

PS i used ladspa, not dmix

[TemplarGR]

TemplarGR, were you beat up as a kid (Or are you still a kid?). Lighten up a little . Its almost Christmas!!! Time of joy and peace....

One thing that Ubuntu has done is make it so most "Normal" people (Like my 75 year old mom) can use Linux without issue. Install and go! How many other Linux distros out there provide this? Yeah, thats what I though... NONE!

Joe

Funny...

I thought someone who states that something doesn't work without any proof would better fit your description about beaten childs...

As for 75 year old Moms, i bet you a million dollars that more of them can use Windows without an issue than Ubuntu... Linux is chosen on technical merits, not ease of use...

[ShadowBane]

"In a typical installation scenario under Linux, the user configures ALSA to use a virtual device provided by PulseAudio. Thus, applications using ALSA will output sound to PulseAudio, which then uses ALSA itself to access the real sound card."

thats one copy that dosent occur with alsa, wouldnt occur at all if pulseaudio was kernel-level

hmmm just been reading about how it was desinged
in theory it should be good, and with mr. Poettering orchestrating it it can achieve what it says

but when i see pulseaudio using 5% of my laptops cpu(could be just ubuntu ppl screwing it up again) to play a low sample rate sound from a youtube video, that tells me its either not there yet or its just doing a lot of useless stuff (5% of a cpu is a whole lot, lot more then people today think)

zero-copy is also a kernel thing that glibc can use, but still the page says that PA has a scheduler running all the time
all in all it can never achieve the efficiency of alsa (alsa with a ladspa plugin for a equaliser used ~1.5cpu to play a 44k mp3 on that shitty laptop)


and for all you talking about sound quality, please dont

Is there something wrong with Pulseaudio's sound quality? The main reason that it uses more CPU time than dmix does for the same job is that uses a better (and more computationally expensive) re-sampler when re-sampling is needed. Another thing that could cause extra cpu usage is doing more re-sampling than needed (setting alternate sample rates in pulse can help with this, if only one thing is running it will use a sample rate that is less expensive to use instead of the primary one.)