SV: [speex-dev] Speex modes
jean-marc.valin at hermes.usherb.ca
Sun Oct 13 16:02:52 PDT 2002
Well, I don't know what SBR is, but there's something in the wideband
mode that may be similar: It's possible to encode the whole 4-8 kHz band
with just ~1-2 kbps by only encoding the (LPC) shape of the spectrum and
then just filling that band with "something that makes sense". Quality
is quite reasonable...
Le dim 13/10/2002 à 06:18, Steve Underwood a écrit :
> Pontus Carlsson wrote:
> >Btw, have you tried using SBR-technology or similar with speech codecs? That
> >might be a good idea I thought.. But I don't know if it produces as good
> >quality with speech codecs as it does for music codecs. Do you know if there
> >is any open source variant of SBR?
> SBR exploits a limitation of your ears. At high frequencies (like over
> 10kHz) you cannot determine pitch with any accuracy. You hear up to
> 15kHz to 20kHz (depending on age and other factors), but you really
> cannot identify pitch at these frequencies. You cannot even determine if
> content above about 10kHz is properly harmonically related to the lower
> pitched fundamentals which usually give rise to them.
> I don't know of any voice specific coder that even attempts to capture
> energy above 10kHz. SBR just isn't relevent. Most wideband speech coding
> captures only 7kHz to 8kHz bandwidth. The key improvement that gives
> over the 3kHz to 4kHz most mainstream voice coders capture is to clean
> up unvoiced sounds. fffff, sssss, and other unvoiced sounds appear
> almost the same at telephone bandwidth. At 7kHz bandwidth they have
> enough character to make them more distinguishable. The basic
> intelligibility improvement you get is usually small. However, the voice
> is rather more pleasant and less tiring to listen to. That brings
> considerable intelligibility improvements in a long discussion. Adding
> energy up to the limit of hearing adds more to the pleasantness of the
> voice, but it isn't usually considered enough to get people excited
> about commiting extra bits per second to it.
> --- >8 ----
> List archives: http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'speex-dev-request at xiph.org'
> containing only the word 'unsubscribe' in the body. No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.
Jean-Marc Valin, M.Sc.A.
Université de Sherbrooke, Québec, Canada
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 242 bytes
Url : http://lists.xiph.org/pipermail/speex-dev/attachments/20021013/ceda047e/signature.pgp
More information about the Speex-dev