AudioEngine branch - DO NOT REQUEST BINARY BUILDS

I noticed that gnif2 copied xbmc into a seperate branch: AudioEngine branch

http://trac.xbmc.org/changeset/32417

Anyone can tell me what's the purpose of this?

To provide an engine for the audio

tslayer Wrote:To provide an engine for the audio

I'm curious about the details. What kind of improvements are planned for example? It the current engine not good enough?

i'm hoping this branch will lead to DRC.

Well, I can

Its a new audioengine to wrap up all the different media types for mixing, samplerate conversion, format conversion, encoding, upmix, downmix, etc...

And yes, DRC is on the tables very early days still, only ALSA works and paplayer is missing alot, but its getting there rather fast. Dont expect ANYTHING to work however, its VERY experimental at this point.

I AM NOT ACCEPTING ANY FEATURE REQUESTS - this is mainly a code clean-up at this point and removal of SDL_Mixer for gui sounds.

Just thought I would post an update to the progress while I am waiting for a build...

So far, SDL has been stripped out of XBMC, a new audio layer called, yes, you guessed it, "AudioEngine" has been built. This new layer is able to handle bit-perfect reproduction of audio, mixing, preloaded sounds, resampling, downmixing, post-processing filters.

What works so far:
* alsa output (NOTHING ELSE)
* dvdplayer
* gui sounds
* paplayer (90%) including x-fade for wsapi users
* visualization add-ons (the interface has not changed, add-ons are compatible)

PAPlayer's stream handing code, cross fade, and gapless mess has been re-written to take advantage of AE. PAPlayer does not re-sample anymore either as AE handles re-sampling, format conversion, channel remapping, down-mixing, making the code in the PAPlayer and dvdplayer MUCH simpler and robust.

Whats still left to do before a release:
* Make paplayer output the codec's native format, currently it converts everything to float
* Make passthrough work again
* Fix paplayer's seek
* Make the other AudioRenderers work (they will probably be re-written somewhat)
* add channel mapping support to paplayer for FLAC files (this one will make lots of people happy)

Obviously NONE of this will make it into Dharma, so don't even ask for it it is way too experimental and un-tested to release just yet.

I know there has been a-lot of stalled "new audio layer" attempts in the past, this time though we are 90% through having everything working where we can move onto new features such as up-mixing, ladspa plugins, etc. There may even be a addon interface for custom post-proc filters at some stage

you mean SDL mixer, SDL for SEMs, window and event handling is still present.

Hi Gnif !

This sounds awesome.

One question just came into my mind for which problem i could not find a theoretical solution:

How can cross-fade handle bit perfect playback in case two sound has to be mixed with different sampling-rate. I think either one stream has to be resampled or both has to be upsampled to a higher bitrate.

Of course this is a bit theoretical because i think most CD material is on 44.1khz/16bit, but there are some 48khz mp3 files out there.

Thanks really much for working on such hard and complex code and yes you are making a lot of people happy with this engine.

Cheers, Alan

obviously you won't get bitperfect anything if you resample/crossfade/adjust volume/do anything. worrying about bit-perfect output with mp3 is funnai though

Thanks for the correction, I did indeed mean SDL Mixer

Yes, spiff is correct here too, however we will try to match the audio devices output sample rate as best we can to the input, but when mixing gets involved, all bets are off.

Here is how the sample rate selection will work (this is not done yet), if there is no streams running, and a GUI sound tries to play, we open at a default rate of 44.1khz

When a stream starts, if is the only one, and the sample rate does not match the currently configured one, AE will re-open the audio output at the correct sample rate to match.

When another stream starts, it is not the only one, so it just gets mixed in to whatever is already open. This will only occur in the paplayer for gapless and cross-fade reasons.

If however passthrough is in use, all streams are re-sampled to 48khz, and encoded after mixing to AC3