AW: AW: [vorbis-dev] PlusV algorithm -> CBR

Tor-Einar Jarnbjo Tor-Einar_Jarnbjo at grosch-link.de
Fri Jan 31 04:56:49 PST 2003



Fredag, 31 januar 2003, skrev du:

<p>>Next thing is, that I would like to clarify the difference between 
a codec's
>latency (which I would die to know in exact figures for the vorbis 
codec one
>day - as well as block sizes, but that's another story) and the 
(bandwidth)
>"smoothing window" latency - call it buffering or whatever.

One thing is the codec latency (which depends on the implementation),
another thing is the latency caused by the file format, which is 
IMHO the interesting number. The file format itself allows block 
lengths between 1 and 32768 samples (in powers of two), but this 
is restricted in Vorbis I to be between 64 and 8192.

The encoder has a minimum latency equal to the block length. With 
a sample rate of 44100Hz, this is between 1.5 and 186ms.

Due to block overlapping, the minimum possible latency in the decoder 
would be 1.5 * max block length. Using only blocks with 64 samples,
you have the "worst case" when computing the 33rd sample of each 
block, for which you need the following block to have been received 
and decoded. With a sample rate of 44100Hz, this is just above 2ms.
Using a block length of 8192, the minimum possible latency would 
be 279ms with the same sample rate.

Adding both values together, you get a minimum total latency of 3.
6ms. In addition to this you have to add an implementation dependent 
value for the encoding and decoding of the packets. A single-threaded 
encoder and decoder fast enough to operate in realtime will not use 
more than 2,9ms for encoding and decoding a 64 sample block.

Tor

<p><p>===================================================================
EASY and FREE access to your email anywhere: http://Mailreader.com/
===================================================================

<p>--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.



More information about the Vorbis-dev mailing list