[CELT-dev] Total Harmonic Distortion THD

Jean-Marc Valin jean-marc.valin at usherbrooke.ca
Wed Aug 18 07:12:25 PDT 2010


On 10-08-18 03:50 AM, Jochen Kilian wrote:
> So the only measure we have is the perceptual audio quality?

Yes. After all, that's what you want in the end, isn't it? SNR and THD 
may be indications of quality when your system is entirely analog, but 
when things get more complex (as they do in codecs), actually listening 
to the audio is the only thing you can do.

> Which are the most recent comparisons you have.
> Looking at the celt_euspico report I wonder if the comparison was made
> with the
> - low complexity version or not
> - fixed point version or not

The comparisons were made with the low-complexity version (I recently 
got rid of the high complexity one anyway) in floating point. Note that 
the fixed-point version should have the same quality as the floating point.

> if not are there major differences in quality for the low complexity and
> the fixed point version?

No (see above).

> Another question I have is if you ever compared the Codec to the APT-X
> codec which is used for Bluetooth wireless headsets.

AFAIK, Bluetooth uses SBC, not APT-X. In any case, CELT is designed for 
lower bit-rates than SBC and APT-X. So using rates like 64 kb/s per 
channel, I would expect CELT to sound better. At very high bitrates 
(>128 kb/s per channel), I have no idea how they would compare since 
CELT is already transparent to my ears. Note that this is based on the 
little information I know of APT-X and SBC since I've never actually 
listened to these. I would encourage you do to some listening for 
yourself to get an opinion.



> Thanks
> Jochen
> 2010/8/18 Jean-Marc Valin <jean-marc.valin at usherbrooke.ca
> <mailto:jean-marc.valin at usherbrooke.ca>>
>     Hi,
>     Measurements like THD (along with SNR and frequency response) belong
>     to the analog world. They are not adequate for describing most forms
>     of digital processing -- and especially not codecs. What you want is
>     perceptual quality tests -- that's all that matters.
>     Cheers,
>             Jean-Marc
>     On 10-08-17 09:26 AM, Jochen Kilian wrote:
>         Hi
>         Has anybody done THD or THD-N measurements with the CELT Codec (bext
>         would be on various bit rates)
>         If someone could share results for Mono at 64kBit and Stereo at 128 and
>         196kBit it would be great.
>         thank you very much
>         Jochen
>         _______________________________________________
>         celt-dev mailing list
>         celt-dev at xiph.org <mailto:celt-dev at xiph.org>
>         http://lists.xiph.org/mailman/listinfo/celt-dev

More information about the celt-dev mailing list