Le Tuesday 21 April 2009 07:09:07 David Baelde, vous avez écrit :
> Hi,
Hi !
> On Tue, Apr 21, 2009 at 4:12 AM, Romain Beauxis <[email protected]>
wrote:
> >> Indeed. I checked, it does not preserve end of tracks and loose
> >> metadata.. ! I will try to fix this soon...
> >
> > Ok, I have done it.
>
> This was indeed why it is marked as hidden and experimental. In my
> mind, fixing it required some mechanisms that would basically amount
> to implement correctly the operators time_warp and lag. Indeed, I
> think that we need to buffer enough data before we can fill a
> resampled frame. Using our existing buffering mechanisms, extended
> with resampling, we could have got metadata and breaks treated
> correctly.
I have looked at this. The problem is that you have to instanciate a fixed
ratio, while in the current implementation the ratio is an interactive float
that can change at any time..
> Your commit is simpler than what I described, but I am not sure that
> it is a (good) solution. You only use one frame as a buffer. If you
> receive one frame of data from the source, and you get less data after
> resampling, you don't have enough to fill a frame, and your #get_frame
> doesn't do anything to compensate that.
>
> One a side note, I'm afraid that the metadata handling might not be
> right: If you use clear instead of advance, the information about past
> metadata is lost, and metadata might be inserted again in the next
> frames.
You"re right, there should be more tests before pulling a frame and I should
change advance instead of clear...
> And.. I can't help saying it: I don't see the interest of packing
> three values in "val mutable data", and it is not very readable -- use
> three instance variables with explanatory names.
>
> > Additionally, I wonder if perhaps we could also pass some informations
> > like this (break temporality) in the new type system that we are
> > planning...
>
> One more time: yes, I know how to do it and will do it in the next
> revision of the type system.
Ok, sorry I didn't understand this before..
Romain
------------------------------------------------------------------------------
Stay on top of everything new and different, both inside and
around Java (TM) technology - register by April 22, and save
$200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco.
300 plus technical and hands-on sessions. Register today.
Use priority code J9JMT32. http://p.sf.net/sfu/p
_______________________________________________
Savonet-devl mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/savonet-devl
