[Vorbis] Sample rates higher than 48k? 24-bit depth?

[Giuliano Pochini]

On 10-Jan-2005 Paul Ellis wrote:
> Giuliano Pochini wrote:
> 
>> I don't know if libvorbis is tuned for 96KHz, but why do you
>> want to use a lossy codec for high quality data ?  Encoding
>> will throw away higher frequencies anyway, so you can down-
>> sample it to 44K before encoding without losing anything.
> 
> I ran some test encodes.  All 24-bit 96kHz input.  Encoded with
> OggDropXPd built from libVorbis 1.1.
> 
> At Q4 Vorbis does cutoff some after 20kHz, however there is still
> information about 20kHz.

-q4 should cut at 18.9KHz


> At Q8 it notches out 20kHz-30kHz for the most part, but above 30kHz
> seems unaffected.

>From -q7 and up it should not apply a lowpass filter anymore.


> Some interesting findings.  At Q4 the 96kHz file was only 200KB larger
> than the downsampled 44kHz file.  Also, even though 740kbit/sec seems
> like a lot, the source file was 235MB and the Vorbis file was only
> 28MB.  About 8.5:1 compression.  Granted it is lossy, but I'd bet a
> 96kHz Q10 Vorbis file sounds better than a 44kHz wav of the same source
> which would be more than 4 times larger.

I don't know. I never did tests with 96KHz tracks. And I don't
think vorbis is tuned for anything other than a few sample rates
(look in libvorbis-1.1.0/lib/modes/) but I'm not sure of that.


> Does anyone know how the input bit-depth really affects Vorbis?  I know
> internal it is all like 48-bit floating point or something to that
> effect (Monty mentioned it earlier in this thread).  Would you really
> attain a higher quality using a 24-bit source instead of a 16-bit source
> though?  If there was a quality difference would it only be noticeable
> at 24-bit?

You can try to encode a 24bits track and its 16bits dowmsapled
version. From the difference of their file sizes you can
roughly evaluate how much information does vorbis keep from
a 24bits source.


--
Giuliano.