[speex-dev] Re: Some simple questions

[Jean-Marc Valin]

> I'm being PHBed into a VOIP project, and Speex sprang to mind.
> Bandwidth is 
> going to be a fairly serious issue for us. With regards to a Speex 
> enc/decoder, I was wondering: Rick Kane and David Siebert have already
> asked 
> about this, but seem to have gotten very different responses - the
> former a 
> call to arms, and the latter a "well, if you do it, it'll get done."
> What's 
> the status of Speex as a fixed point codec? Is anything done at all? 

Fixed-point is one of the main thing we'd like to get for Speex.
However, if it is to happen someone needs to either 1) write it or 2)
pay someone to write it (if many people are interested in 2, that may
make it cheaper).

> The homepage suggests that Speex is appropriate for VOIP. It seems
> that 
> most of the iPaqs, Intermecs, and Symbols (Your FAQ explicitly
> mentions 
> iPaqs as slow - ack!) use ~200mHz or ~400mHz ARM CPUs. Is that enough
> for 
> realtime encoding *while* decoding? If so, is there an estimate of how
> much 
> free CPU time will remain? Enough for other simultaneous 
> compression/decompression? How much effect would a fixed-point encoder
> have 
> on performance? Should I start considering commercial codecs, such as 
> Nellymoser's SASE and ASAO? Are there other free or commercial codecs
> I 
> should be looking at? Is there maybe an analog to the comparison chart
> at 
> the website that gives performance characteristics, etc? Or, maybe,
> even 
> just an idea of the baseline CPU that's realistic for realtime full
> duplex 
> VOIP? 

First, note that decoding is much less complex than encoding, so let's
just consider encoding. For 8 kbps, Speex requires about 10 mflops when
set to complexity 1 (minimum complexity). With float emulation, this is
too slow for an iPaq, but with a fixed-point, it would probably fit
easily. I can't say much more because 1) there's no fixed-point yet and
2) I don't know exactly how fast the ARM is.

> Also, IANAL. The xiph.org variant of the BSD license looks *really*
> simple, 
> but I'd like to check and be sure. If I were to use Speex in a closed 
> source binary-only commercial product, it would be sufficient to note
> the 
> copyright and the list of conditions in the manual, yes? (I also
> intend to 
> put the copyright notice in the about box, but since it's a handheld I
> wasn't going to give up screen real estate to the stipulations.)

IANAL, but the way I see it is:
1) Don't claim you wrote it and put the copyright notice somewhere (box,
manual, ...)
2) Don't say "Xiph.org thinks our product is great" without
authorization.

Otherwise, you can do pretty much what you want with Speex.

        Jean-Marc


-- 
Jean-Marc Valin, M.Sc.A., ing. jr.
LABORIUS (http://www.gel.usherb.ca/laborius)
Université de Sherbrooke, Québec, Canada


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 190 bytes
Desc: Ceci est une partie de message numériquement signée
Url : http://lists.xiph.org/pipermail/speex-dev/attachments/20030919/0d0f9f65/signature-0001.pgp