[tremor] [Tremor - Decoder Compliance]

[Johannes Sandvall]

> I am new to vorbis development. I am trying to port Vorbis integer only 
> decoder Tremor to the STMicroelectronics' VLIW ST200 series of fixed 
> point processors.
> 
> My concerns are :
> 
> 1/ Is starting from the Tremor integer only codec a good idea or should 
> I target the floating point code of Vorbis?
>    My target processor is a fixed point one.

Yes. Consider the low-mem branch if memory is an issue.

> 2/ What are the tools/test vectors/certification criteria that are 
> needed for the Vorbis Compliance of the Decoder that I will port?
>    In particular where can I get a rigorous test suite to check the 
> compliance of my fixed point vorbis decoder complaince?

There is none! I do not fully understand why there have not been any
effort creating this. The easiest way to a functional test of the
decoder is to compare the output from your decoder with the output
from the floating point implementation. Is then easy to develope a
matlab script to test for maximum deviance and RMS. A resonable
criteria can be found in the MP3 compliance documentation. 

What would be nice though is synthetic test vectors and decoded songs
for different configurations (e.g. window length, codebooks). A few
other decoders offers synthetic decoded songs that tests e.g. all
possible cases of the huffman packing. That would be nice to have.

If thats tricky a simple collection of songs encoded with different
window length and different sets of codebooks would be nice. Even
though if someone provided the list with such a collection it would
probably take years before they would be available from
xiph.org... (I'm not sure that the over a year old critical memory leak
patches that were sent to the list is merged to the repository...) 

/Johannes Sandvall
--- >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 'tremor-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.