[Speex-dev] PESQ results for speex 1.0.3

[Steve Underwood]

Ben Greear wrote:

> Francois Menard (Mailing List Account) wrote:
>
>>
>> did you try speex in wideband mode ... what bitrate?
>>
>> f.
>
>
> 15kbps mode.  It does significantly better with a different
> speaker (the male does better than the female), as well.

Most CELP based codecs do a better job with lower pitched voices, so 
male voices tend to sound better than female voices.

> I'm considering purchasing a commercial PESQ license and wrap
> the PESQ software in a server application.  I could then allow
> other folks to run PESQ comparisons on my server through some API,
> either for free or for a fee.  (ie, commercial customers for $$,
> open source & hoby-ists for free)...

I have mixed feelings about the value of that. People are starting to 
trust PESQ scores, like they are a solid measure of how a human would 
perceive the quality. That could mean codecs will win in the marketplace 
if they sound bad, but score well. PESQ is pretty useful for quick 
checkouts of system setups. It is a pretty lousy way to judge the worth 
of a codec.

> I'm hoping this might allow Speex & similar developers test their
> codecs in some standardized manner without having to pay the thousands
> of dollars for a PESQ license...

PESQ is being pushed hard to people with limited knowledge of the 
subject, and the superficially useful results it produces sells 
licences. Beware of the true value of things which make money. Snake oil 
makes money. :-)

Regards,
Steve