[Asterisk-Dev] results: Intel's IPP G.729 with Asterisk

[Adam Hart]

Daniel Pocock wrote:
> I've ported Intel's IPP-based codec to Asterisk and the results are 
> printed below.  Does anyone know how this compares to Digium's codec - 
> could someone post similar stats?
> 
> The code is currently pre-alpha - the Intel demo didn't look after 
> releasing memory, etc, because it was just a demo to be run in 
> `one-shot' mode.  If the feedback I get suggests that these benchmarks 
> are good, then I will clean up my implementation so that it is safe for 
> continuous, 24x7 operation, and if messy patent and copyright laws 
> permit, release it publicly.  If anyone believes this is worth pursuing 
> further, please feel free to contact me on or off the list.
> 
> This was executed on a P4 2.8GHz with 512MB RAM.  Lot's of Mozilla and 
> virtual machine based applications running at the same time too.  The 
> phone I'm using with it is a Budgetone 102 with latest firmware, audio 
> sounds fine in both directions.
> 
> h1*CLI> show translation
>        Translation times between formats (in milliseconds)
>         Source Format (Rows) Destination Format(Columns)
> 
>        G723   GSM  ULAW  ALAW  G726 ADPCM SLINR LPC10 G729A SPEEX  ILBC
>  G723     -     -     -     -     -     -     -     -     -     -     -
>   GSM     -     -     2     2     -     2     1     8    12     -    24
>  ULAW     -     6     -     1     -     2     1     8    12     -    24
>  ALAW     -     6     1     -     -     2     1     8    12     -    24
>  G726     -     -     -     -     -     -     -     -     -     -     -
> ADPCM     -     6     2     2     -     -     1     8    12     -    24
> SLINR     -     5     1     1     -     1     -     7    11     -    23
> LPC10     -     7     3     3     -     3     2     -    13     -    25
> G729A     -     7     3     3     -     3     2     9     -     -    25
> SPEEX     -     -     -     -     -     -     -     -     -     -     -
>  ILBC     -     9     5     5     -     5     4    11    15     -     -

Looks like your P4 was under load as most thoses numbers look double, 
here's a P4 3.0

          G723   GSM  ULAW  ALAW  G726 ADPCM SLINR LPC10 G729A SPEEX ILBC
   G723     -     -     -     -     -     -     -     -     -     -     -
    GSM     -     -     2     2     2     2     1     5     -    23    13
   ULAW     -     3     -     1     2     2     1     5     -    23    13
   ALAW     -     3     1     -     2     2     1     5     -    23    13
   G726     -     3     2     2     -     2     1     5     -    23    13
   ADPCM    -     3     2     2     2     -     1     5     -    23    13
  SLINR     -     2     1     1     1     1     -     4     -    22    12
  LPC10     -     4     3     3     3     3     2     -     -    24    14
  G729A     -     -     -     -     -     -     -     -     -     -     -
  SPEEX     -     3     2     2     2     2     1     5     -     -    13
   ILBC     -     4     3     3     3     3     2     6     -    24     -

I'd say intel's G.729 would run around 7ms on this PC given your numbers 
above - Digium's implementation runs around 11ms

-Adam