[MPlayer-users] introducing DVD-to-DivX approach that fixes A/V sync problems

[Rich Felker]

On Wed, Jul 06, 2005 at 12:29:40PM +0200, Joachim Jautz wrote:
> On Tue, Jul 05, 2005 at 01:33:03PM -0400, Rich Felker wrote:
> > 1, using frameno.avi. STRONGLY discouraged. it will not work with any
> > inverse telecine, frame decimation, etc., which are needed for most
> > movies.
> 
> Okay, Martin Collins pointed that out, too. I do not stick to that
> frameno-method if there is a better way.
> 
> But, let's look at it from a practical point of view. The problem
> is that using the frameno.avi _seems_ to be the only way to
> achieve these goals:
> 
> - the video is to be encoded only once (in two passes, of course)
> - multiple audio tracks (e.g. `en' and `de') are desired; moreover
>   it would be nice if these audio tracks were
>   * MP3 encoded (instead of -oac copy)
>   * `normalized' or in other words: not so silent that you have to
>     pump up your stereo to volume `blow it' ;)
> - each audio track is to be mixed to the video resulting in one
>   stand-alone movie file
> - target format is MPEG4 / DivX5
> 
> My assumption is that the recommended way that is described in the
> mplayer documentation requires to encode the video twice if you
> want to get two different language versions.

Try this:

dump audio to fifo with: mplayer -ao pcm ...
encode from fifo with: lame ...
(repeat for each audio track)
encode video pass 1, with: mencoder -audiofile lang1.mp3 -oac copy ...
encode video pass 2, with: mencoder -audiofile lang1.mp3 -oac copy ...
use whatever program you like to mux all the other audio tracks in too

Even for just one audio track, this is the only clean way I know to
avoid a/v desync with mencoder..

> > 2, 'normalizing' audio. if you normalize to maximize peak, you WILL
> > generate SERIOUS distortion when encoding with any frequency-domain
> > audio codec (i.e. any decent non-lossless codec). the headroom away
> > from peak is needed to prevent clipping in the transformations. you
> > should either normalize rms instead (to a moderate value, not high),
> > or do no normalization at all!
> 
> I'm sure there are scenarios in which your warning is good advice
> and has to be considered.

No, "there are scenarios" is incorrect. PSNR is ALWAYS significantly
lower, and distortions can be heard, if you peak-normalize then encode
to mp3, vorbis, etc.

> Regretfully I cannot see why there should occur distortion when
> - dumping audio to a wave file
> - adjusting the file so that its maximum sample is at full scale
>   which just gives a file the maximum volume possible without
>   clipping (cf. [1])
> - encoding the wave file into MPEG Layer 3
> 
> Do I miss some important technical detail?

Yes, the one I described above.

> If the distortion were
> `SERIOUS', it would be audible, I assume? To me the result sounds
> perfect.

Your ears are bad. Don't feel bad about it, mine are too. But people
with good ears will writhe in pain listening to your bad encodes.

Rich