[Speex-dev] To CELP or not to CELP ... at higher bitrates

Jean-Marc Valin Jean-Marc.Valin at USherbrooke.ca
Sat Oct 29 05:55:40 PDT 2005

> My approach, therefore, is to build a Windows "wave" file asynchronously by 
> using a streaming retransmission protocol to request the retransmission of 
> any lost packets from a short-term history buffer maintained by each 
> sender, thus "filling in any gaps" after the fact.  In this way we can 
> guarantee that a perfect recording will always result.

For *near*-perfect recording, record at 96 kHz/24 bps, 8 channels,
uncompressed. That'll be about 18 Mbps bitrate. 

> Since the resulting recorded quality is the ONLY concern, and since we can 
> stipulate that all parties will always have access to broadband-scale 
> connectivity, does using a CELP-based codec such as Speex make the most 
> sense in this application?

You already answered the question, haven't you? ;-)

> I would be running Speex in ultra-wideband with 32 or 48 kHz sampling and 
> its bitrate completely open-ended and upwards of 44 kbps ... but higher bit 
> rates (of several hundred kbps) would also be readily available to this 
> application.

I don't recommend ultra-wideband if you require high-quality.

> Given that we really don't need the compression levels offered by advanced 
> CELP speech encoding, does it still make the most sense to use Speex, or 
> would we be better served to use some other codec -- perhaps such as mp2 or 
> mp3 -- at higher bitrates?
> Do you have any guidelines you could share?

Use FLAC or Vorbis.


More information about the Speex-dev mailing list