You my wish to read my article at www.ruccas.org. 

The difference that I heard was obvious only in one section of one piece. In all other places, I could for the life of me and despite heavy listening, not hear anything different -- even though I thought I could hear differences.  In other words, in a number of places I thought I heard differences, but the test software said I was only really hearing a difference in one place.

So you may think you hear a difference, but maybe you are not really hearing a difference. 

The tests I did were double blind. I had two sources for the exact same piece, different only in that one was rendered with 64 bit Csound and the other was rendered with 32 bit Csound.  I would pick a section of the piece. The software would pick which source (32 or 64 bit) at random. Then it would ask me to guess which source it was. This would be repeated a dozen or so times and the binomial distribution gives the odds my guesses were random or better than random. For that one section of one piece, the odds were extremely good that my guesses were not random. In all the other sections of the other pieces, my guesses were no better than random.

It's enough for me that one section was clearly different, as that is sufficient to show that 64 bit Csound is more accurate than 32 bit. I presume that with more testing, I would find more sections of more pieces with audible differences.

It would be interesting to do more tests, with more listeners and more pieces and more sound sources, but I don't have time for that now.

I also am confident that some types of software instruments would be more sensitive to precision than other types.


Regards,
Mike

-----Original Message-----
From: David Akbari 
Sent: Sep 1, 2005 8:30 PM
To: csound@lists.bath.ac.uk
Subject: Re: [Csnd] single vs. double (was: Conversion into 48 khz and 24 bit, does it make sense?)

You know, since you've mentioned it both here and on the music-dsp 
lists, I've started to listen more carefully to sounds rendered with 
single and double precision and I really think you're on to something 
with that idea.

Double precision definitely sounds better... at present are there any 
studies and/or papers that deal with the subject more in depth ?? I'm 
quite interested to learn more about these phenomena.


-David

On Sep 1, 2005, at 6:01 PM, Michael Gogins wrote:

> At www.ruccas.org, I have posted results of double-blind listening 
> tests comparing Csound compiled for 32 bit samples and compiled for 64 
> bit samples. In one case there was a clearly audible different.
>
> In both cases, soundfiles were 96 KHz float stereo.
>
> This is not directly relevant to your case, but I think it indicates 
> that in some types of sounds, there will be differences that careful 
> listeners can hear.
>
> Regards,
> Mike
>
> -----Original Message-----
> From: Jan Jacob Hofmann 
> Sent: Sep 1, 2005 12:45 PM
> To: Csound List 
> Subject: [Csnd] Conversion into 48 khz and 24 bit, does it make sense?
>
> Dear list,
>
> I am planing to do a major remix of my pieces. They consist of
> soundfiles originally recorded at a sample-rate of 44,1 khz at 16 bit.
> I wonder if there would be an increase of sound-quality if I converted
> these files into 48 khz and 24 bit beforehand. I know the files
> themselves would surely not sound better themselves, but as
> reverberation and early reflections are added in the course of he mix,
> aswell as the amplitude of these files is altered, I guess doing it in
> 48 khz and 24 bit might be an advantage: the higher sample-rate would
> give a better temporal resolution (important for the early reflections)
> and the higher bit-rate more definition and a higher headroom for the
> amplitude level. What do other Csounders think? Do my thoughts on this
> make sense?
>
> Best regards,
>
> Jan Jacob

-- 
Send bugs reports to this list.
To unsubscribe, send email to csound-unsubscribe@lists.bath.ac.uk



-- 
Send bugs reports to this list.
To unsubscribe, send email to csound-unsubscribe@lists.bath.ac.uk