[Speex-dev] CELT 0.3.2, listening tests

Jean-Marc Valin jean-marc.valin at usherbrooke.ca
Mon May 19 15:26:13 PDT 2008

Alain M. a écrit :
> I was very impressed by the comparative results. Could you give a bit 
> more information about what really are 7kHz and 3.5kHz in the 48kHz table?

They're anchors. The 3.5 kHz anchor is a low-pass filter with a cutoff
of 3.5 kHz. So it's effectively equivalent to the best quality you can
have with a narrowband (8kHz sampling rate) signal. Similarly, the 7 kHz
anchor is the best quality you can do with a wideband signal (16 kHz
sampling rate).

> I am looking forward that this will bring a bigger separation between 
> the basic tools and speex in a sense that the basic tools could be used 
> with more than one codec. I even understand that there is interest in 
> the tools alone (Echo supression, ressampler, Jitter buffer, VAD,...)

Well, since 1.2beta3, the "tools" are already in their own library,



> Alain
> Jean-Marc Valin escreveu:
>> Hello all,
>> This is slightly off-topic, but should be of interest to some people on
>> this list. I just released version 0.3.2 of the CELT ultra low-delay
>> audio codec (http://www.celt-codec.org/). CELT is designed to encode
>> high quality speech and music with less than 10 ms delay and at rates
>> starting from around 32 kbit/s.
>> This version is "special" in that it is the basis for some listening
>> tests I conducted for an IEEE paper. The results of these tests are
>> included in the comparison page: http://www.celt-codec.org/comparison/
>> along with some audio samples at http://www.celt-codec.org/samples/tasl/
>> Cheers,
>> 	Jean-Marc
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev

More information about the Speex-dev mailing list