Slashdot Mirror
Speex Joins Xiph To Bring Free VOIP To The Masses
xercist writes "Xiph.org
has added a new project to their plate of goodies-
Speex.
Speex is an audio codec specifically for, you guessed it, voice.
It has integration with Xiph's
OGG
container, but is mainly being used right now for VOIP.
There is currently an XMMS
plugin
available, and is also supported by
LinPhone,
OpenH323,
and
GnomeMeeting.
Asterisk PBX
is working on adding support.
This is not a new project -- Jean-Marc Valin has been hard at work writing
the codec for quite a while now. However, Jean-Marc is now a full-fledged
member or the Xiph.org team, and in celebration, Speex beta one is being
released.
Xiph.org has brought you
(or is currently working on bringing you)
Vorbis,
Tremor,
Theora,
Tarkin,
Icecast2,
cdparanoia,
now Speex,
and, of course, the
Moaning Goat Meter.
This is a LOT to do, so please
donate
to show your support."
Why does everyone insist we need to do absoultely everything over TCP/IP? We already have this interesting thing called the telephone network that works quite well at this very thing. Sorry for the trollish retort, but it seems like we should solve other very real *problems* with software, instead of going out of the way to propose yet another trendy TCP/IP enabled whatever.
I didn't even know that Monty had also done cdparanoia until recently. Great coder.
Here's to Xiph -- singlehandledly taking on the tech-media companies (Real/Apple/MS and tons of failed companies) and steadily gaining ground.
We've had propriatary media formats for a long time. (Incidently, propriatary file formats are one of the strongest weapons incumbents have against upstart open source projects). This is a big movement that's starting to cascade, with more companies joining the Xiph bandwagon daily (and little interest in the MPEG4 people).
May we never see th
I already donated US$7.50 to xiph (10 times the price of an mp3 decoder). It's really not much, but it seemed about right in proportion to the other software I use (especially since I only use xiph's products for my desktop, not servers). I know xiph is working hard, but I feel more of a debt to the creators of linux, gnu, postgres, apache, exim, debian, python, perl, openbsd (I only use their openssh, but that's important), and all the other great projects I didn't mention (and those aren't necessarily in order).
Social scientists are inspired by theories; scientists are humbled by facts.
The badly dated Half-Life engine still has one feature that is more powerful than anything in UT2k3: and that's voice over ip, in-game, mostly used in Counter-Strike.
UT2k3 already uses OGG for its music -- and I recall reading a UT2k3 developer plan file that states the wish for voice-over-ip, but basically they were waiting for someone in the open-source world to do all the work.
Why just hit a few buttons to say, "Ownage!" when they can hear your true compressed, overly nasal-sounding voice say it -- or perhaps more insulting, filthier things?
It's a simple api, and the audio quality comes out okay for voice too. (unless you try sending music through, then it really just craps out)
It is indeed really easy to use. But one thing I've noticed is the encoder uses a *LOT* of CPU time. My P3/800 was maxed out while encoding.
Perhaps this is just something I've done. What kinda processing power did your own project need? (While encoding)
Yes!
Generally voice is sampled at 8kHz because it generally ranges from around 200Hz to 4kHz and Nyquiest suggests we need a digital sampling rate of twice the highest frequency.
The reason music sounds crappy is that in a well designed system you will loose all the audio components above 4kHz, and in a badly designed system they will manifest themselves as other lower frequencies.
Now this *does not* mean that this codec is a bad codec - merely that's it's one optimised for a specific task.
[one of the IETF's iLBC draft authors - just to avoid confusion that might be caused for people that are not yet familiar with iLBC]
... and where iLBC scores better (for those PL values as well) then other CELP based coders in tests done by recognized independent labs (Dynastat) ...
>They give access to their patents as long as you're using it in iLBC.
Right, and iLBC is an IETF draft = anyone can contribute to it and use it. Getting the freeware codec standard this way, it is easier to achieve outcome result being royalty free, which IMHO I find very grim for anything what is CELP based (+600 patents associated with CELP - source of info = delphion patent dbase, colleagues coauthors of the draft with +50 years of speech coding experience - authors/coauthors of a number of existing speech coding stds).
>but for reasonnable operation (2-5%), the Speex quality is (to my ear better).
Interesting observation. So far, we were getting from different users/communities contrary results, which IMHO I find coherent due to CELP's inter-packet memory dependency, where when losing one packet You are losing properties of the packets that are following, propagating error
>CELP patent is expired and I have been careful not to include things like ACELP and other patented algorithms.
Please see above.
Best regards,
Alan Duric