[gstreamer-bugs] [Bug 590363] a52dec negotiates wrong channel downmix

[GStreamer (bugzilla.gnome.org)]

If you have any questions why you received this email, please see the text at
the end of this email. Replies to this email are NOT read, please see the text
at the end of this email. You can add comments to this bug at:
  http://bugzilla.gnome.org/show_bug.cgi?id=590363

  GStreamer | gst-plugins-ugly | Ver: git

Sebastian Dröge changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |slomo at circular-chaos.org




------- Comment #1 from Sebastian Dröge  2009-08-14 08:29 UTC -------
(In reply to comment #0)
> a52dec picks first structure in the caps allowed downstream and uses it.
> Instead, a52dec should
> 1) Search for a structure that matches its natural channel count and channel
> layout exactly and (if it is found) output audio as-is.
> 2) if 1) fails, search for a structure that matches its natural channel count
> (but not layout) and (if it is found) output audio with this channel count,
> (moving its channels to match the layout in the caps?)
> 3) if 2) fails, search for a structure that does have channel count less than
> its natural channel count and output that channel count (with downmixing?).
> 4) If 3) fails, search for a structure that does have the smaller channel count
> and ... upmix? Or maybe just use its layout to output native channel count?

Step 2) won't work always as a52dec can't convert to every possible channel
layout. For example it can't convert to (REAR_LEFT,REAR_RIGHT). Not that this
is a useful channel layout but it would be a problem ;)
In step 2 you could check if there's a structure with the natural channel count
and a permutation of the natural channel layout.

Step 3) should be more a search for a structure with higher channel count and
"compatible" channel layout, i.e. one that a52dec can convert to. Next step
would be the same with lower channel count.

And if all that fails return NOT_NEGOTIATED ;)

Also note, that if downstream doesn't specify a channel layout but only a
channel count this means that you can provide whatever channel layout you want
(i.e. downstream doesn't care at all what the channel layout is, the volume
element for example).

> P.S. This is really confusing, and somewhat wrong. libavcodec, for example,
> defines special stereo downmix layout, which is used to tell the decoder to
> downmix more-than-2-channel-audio to stereo instead of picking front-left and
> front-right channels and pushing them as-is. For example, there could be a
> 6-channel audio (left, right, center, lfe, rear left, rear right) and a
> 5-channel downstream caps with channel-positions="left, right, center, rear
> left, rear right". It is ambiguous. Should the decoder downmix 6->5 (in this
> case - mix LFE into all other channels) or just pick 5 channels (output 5
> channels as is and discard LFE)? This is not addressed by current channel
> layout signaling system.

a52dec has also special N->2 downmix matrices and they're used when downmixing
in a52dec IIRC. In your 6->5 example IMHO a52dec should use it's internal
downmixing matrix to downmix to the 5 channels, if such thing doesn't exist it
should probably just drop LFE.

> P.P.S. Situation is worsened by the fact that audioconvert (which is usually
> placed after any audio decoder) does not know how to convey downstream caps
> exactly and discards channel layout information (and often merges channel count
> information). Because of that it is not possible to use it for channel
> reordering (as in 3) and it is not possible to negotiate channel layout (as in
> 2) when audioconvert is placed after the decoder.

audioconvert tries to pass the channel layout downstream as is if possible,
then it tries to pass a permutation of the channel layout downstream and if
everything fails it does magic mixing (magic because it has no special
knowledge about the channels except their positions, i.e. it can't use a
special 5.1->2 downmix matrix as should be used for MPEG for example).

If audioconvert is after a52dec, what caps do you get? You should probably get
the downstream caps as the first structures and then the "audio ANY caps"
afterwards. In that case you can still do you special handling in a52dec.
If this is not the case audioconvert should be fixed I guess ;)

Do you want to work on this? :)


-- 
See http://bugzilla.gnome.org/page.cgi?id=email.html for more info about why you received
this email, why you can't respond via email, how to stop receiving
emails (or reduce the number you receive), and how to contact someone
if you are having problems with the system.

You can add comments to this bug at http://bugzilla.gnome.org/show_bug.cgi?id=590363.