Hi there I have a few questions regarding some technical aspects of Vorbis(part of a course I have to resit unfortuantely). These weren't very well documented in my lecture material so I will try and express my understanding of the processes.
One issue I am having is having a technical explanation for Vorbis being an asymmetric codec.
I know that an asymmetric codec is a codec that takes longer to encode than to decode. Vorbis has a complex coder that is based on psychoacoustics and a simple fast decoder. Can anyone provide me with links to technical information on these parts of the coding process? I have been looking but to no avail.
The windowing functions in Vorbis can also be asymmetrical. Is this because of the MDCT's lapping properties? I have also read that this is also due to the the variation in the block size as well. The block size can vary from frame to frame, this can result in the window being asymmetrical with the shorter window determining the number of samples falling under the window?
if anyone can correct me on these aspects or provide me with links to technical documentation. I would be very grateful!
Hi,
most of your questions should be answered by the Vorbis I specification (http://www.xiph.org/vorbis/doc/Vorbis_I_spec.html), although this may, in part, be quite difficult to understand.
For encoder specifics you need to take a look at the source code of the libvorbis reference encoder (http://downloads.xiph.org/releases/vorbis/).
Yes, the windows are asymmetrical if you are switching between different transform lengths (= block sizes). Same as in AAC.
Chris
Thanks!
The windowing functions in Vorbis can also be asymmetrical. Is this because of the MDCT's lapping properties? I have also read that this is also due to the the variation in the block size as well. The block size can vary from frame to frame, this can result in the window being asymmetrical with the shorter window determining the number of samples falling under the window?
Yes this is true. The block sizes vary for different sampling rates. By default the block sizes used are 2048 samples on long blocks and 256 samples on short blocks for 44.1 kHz the default sampling rate. I think the asymmetrical nature in the block sizes you are referring to are called "transition blocks" and yes it is, because of the MDCT overlapping properties. There used to be a research paper floating around on the web that visually depicted how it worked theoretically. It has since vanished though. Sorry . You might find the paper below useful from a mathematical standpoint.
MDCT in Vorbis:
http://www.free-comp-shop.com/vorbis.pdf (http://www.free-comp-shop.com/vorbis.pdf)