Vorbis dev - Aug 2005 - Input buffer requirement.

Hi all,

I am in the thinking phase of Vorbis 'file' to 'buffer'
conversion.
ie, accessing the stream data from the static buffer, rather than
accessing the file from the decoder( the way it is implemented in the
vorbis decoder).

I am putting my understanding and the respective query below,

As the general convention , there are two methods by which i can do this.

1) Assume what ever size input buffer.
-------------------------------------------------------
    In this case,  from every point in the decoder when ever the input
stream is accessed, it has to check for wheather the end of the buffer
has reached or not.\

If (end of buffer )
{
then request for populating the input buffer;
}
else
{
continue processing;
}

2) Assume a max input buffer size.
---------------------------------------------------
     In this case, every time the decoder is invoked we are sure the
input buffer contains full data required for that audio packet.
So, there is no need for any complexity inside the decoder like
checking for end of buffer or so.

But, vorbis specification is not giving any max limit (or minimum
compression) figure. (In AAC case the standard mention about this
scenario)

What is the input buffer size which i can assume, which will cover
real world "OGG VORBIS" stream. ie none of the input granule won't
exceed the max limit.

What is the choice of the implementors ??

I found it in one source that, 
they have assumed 8KB input buffer. they are saying they hav't come
accross any ogg stream which exceed this limit.

How risky is this assumption ?

Can I go ahead with this figure ?? 

- sher

Hi all, 

To add to the input buffer requirement.... 
I did some calculation ... 

If our Ogg Vorbis codec configuration is,

Bitrate = ABR up to 496 kbps (per channel).

max channels = 2 ch

Max sampling frequency = 48 KHz

 

Considering the worst case vector, 48 KHz 2 - channel.

Input stram bit rate is 496*2 = 992 kbps (kbits per sec).

992 kbps = 992/8 kBps = 124 kBps (kBytes per sec)

To represent '1 sec' of data (represents 48000 samples in time), it
requires worst case *128 kB*.

 

If 4096 is the *max* samples / frame / channel as per Vorbis** (Is
*4096* correct???)

Then the input packet size is nearly (124/48000)*4096 ~= 10.581 KB  

Is this calculation correct ??? or near to the real world streams ??? 



On 8/1/05, Sherin S <sherin.s@gmail.com> wrote:
> Hi all,
> 
> I am in the thinking phase of Vorbis 'file' to 'buffer'
conversion.
> ie, accessing the stream data from the static buffer, rather than
> accessing the file from the decoder( the way it is implemented in the
> vorbis decoder).
> 
> I am putting my understanding and the respective query below,
> 
> As the general convention , there are two methods by which i can do this.
> 
> 1) Assume what ever size input buffer.
> -------------------------------------------------------
>    In this case,  from every point in the decoder when ever the input
> stream is accessed, it has to check for wheather the end of the buffer
> has reached or not.\
> 
> If (end of buffer )
> {
> then request for populating the input buffer;
> }
> else
> {
> continue processing;
> }
> 
> 2) Assume a max input buffer size.
> ---------------------------------------------------
>     In this case, every time the decoder is invoked we are sure the
> input buffer contains full data required for that audio packet.
> So, there is no need for any complexity inside the decoder like
> checking for end of buffer or so.
> 
> But, vorbis specification is not giving any max limit (or minimum
> compression) figure. (In AAC case the standard mention about this
> scenario)
> 
> What is the input buffer size which i can assume, which will cover
> real world "OGG VORBIS" stream. ie none of the input granule
won't
> exceed the max limit.
> 
> What is the choice of the implementors ??
> 
> I found it in one source that,
> they have assumed 8KB input buffer. they are saying they hav't come
> accross any ogg stream which exceed this limit.
> 
> How risky is this assumption ?
> 
> Can I go ahead with this figure ??
> 
> - sher
>

Hi!

The theoretical maximal size of a stereo vorbis packet is over 256kbytes.  For 
real world streams the 8kb in the source is plenty.

bye
Denes

h?tf? 01 augusztus 2005 16.51-kor Sherin S ezeket a bolcs gondolatokat 
fogalmazta meg:
> Then the input packet size is nearly (124/48000)*4096 ~= 10.581 KB
>
> Is this calculation correct ??? or near to the real world streams ???
>
-- 
- Use the Source Luke ! -

Hi,
Are there any published papers that discuss the vorbis encoder in 
detail. I searched IEEE and google, but could not find any (possibly 
published under some other name).
Vorbis -I spec document does give a lot of detail, but I am looking for 
a more generic overivew.

Any references are appreciated.
Shank

Thanks in advance for any help.

I am a bit confused with the number of codebooks that are used in 
vorbis. Why are multiple codebooks necessary ?(in one music file, I 
found around 42 codebooks).

Shank