[CELT-dev] celt vs speex

Jean-Marc Valin jean-marc.valin at usherbrooke.ca
Mon Oct 18 17:13:54 PDT 2010

On 10-10-18 02:40 PM, Pascal Pochol wrote:
> Wow. hum I don't want to crowed the celt-dev list with Harmony question is
> there an harmony-dev list somewhere?

Actually, the IETF now has its (hopefully) final name, which is "Opus".
The associated mailing list is codec at ietf.org .

> What about cpu wise for the decoder? We had to limit ourselves to 16Khz
> speex fixed point decoding because 32Khz was taking too much cpu. We do the
> encoding on PC machine so no problem there but the decoding in the worst
> case is done on a tiny ARM9 device.

You can try for yourself, but I wouldn't be surprised if Opus/CELT at 48
kHz was lower complexity than Speex at 32 kHz -- maybe simiar to Speex
at 16 kHz, but I haven't tested.

> How production ready is harmony assuming that the stream and the code don't
> need to change once I release the application? I assume that since it's
> built upon speex and celt it's already as good as both? But I've been wrong
> in my assumption in the past.

Opus is build on SILK and CELT -- it has nothing to do with Speex. In
fact it's better than Speex across the board. As for the bit-stream,
we're working on freezing it, but it's not done yet.


More information about the celt-dev mailing list