[linux-audio-user] Re: Digital Fidelity

Dan Mills dmills at spamblock.demon.co.uk
Mon Feb 27 14:16:29 EST 2006

On 27 Feb 2006, at 18:11, Maluvia wrote:

> Apparently digital fidelity is a non-issue around here.
> If you guys really feel that you can produce professional-sounding,
> commercial quality CDs with 16-bit sound cards and 'correct'  
> dithering - be
> my guest, and good luck.
> Music that originates in the digital domain doesn't even need to be
> sampled.
> Perhaps there are many here just doing all this for fun or as a  
> hobby - I
> have no idea.
> I am guessing that there are not many people on this list recording
> acoustic instruments or classical-type music.
> I doubt that a recording engineer trying to record a violin, harp or
> orchestra, would be happy using a 16-bit sound card.

There are two separate issues here, one being a suitable word length  
for recording, the other being a suitable word length for distribution.

For distribution, I would contend that 16 bits is fine (and in fact  
that the vast majority of CDs don't come close to using that). After  
all by this point in the process we know what the peak value is.....

For recording (where you want 20db of headroom above nominal level to  
avoid clipping that once in a lifetime take, and where you don't  
really know what the actual peak levels will be, then sure go to 18  
or 20 or 24 bit (and accept that the low bits will be essentially  
random). This is not because you are likely to have >96db dynamic  
range for most instruments but because you don't know ahead of time   
where that range starts and stops.

> There *is* generational loss in the digital domain, as well as in the
> analog domain.
> Sorry.
> If you think the subtle differences I hear are auditory  
> hallucination or
> self-hypnosis, and you can't hear them yourself - I can't convince you
> otherwise - I see no reason to even try.

I don't buy it, sure I buy jitter being a issue at the DAC, but that  
is ONLY an issue with conversion to or from analogue, it cannot be an  
issue with the storage or with digital transfer between storage  
devices where either the data recovery is OK is it is not. If the MD5  
hashes match the data is probably identical.....

> But there is a big difference between saying that this loss is  
> negligible
> and insignificant, and saying it simply does not exist.
> There are some physical laws working against your premise here but  
> I won't
> go into them - I don't want to take this thread OT yet again and  
> turn it
> into a discourse on physics.

If the diff returns nothing then the files are identical, at least as  
far as the computer is concerned (or diff has a bug). Sure hard disks  
have a specified bit error rate, but it is fairly close to zero, as  
the fact that these things work at all can attest.

> There are others - professional audio engineers - who also hear  
> these kind
> of differences, but I guess they must all be into metaphysics,  
> hocus-pocus
> and self-delusion as well. (A lot of money in that.)
> Here are a few articles that touch on this subject, and say a lot  
> of what I
> have been saying:
> http://www.johnvestman.com/digital_myth.htm
> http://www.johnvestman.com/digital_myth2.htm

Seems to me to be saying that as long as you actually manage to make  
an exact copy of the file, the only issue is  jitter going to the  
DAC? Checking you have an exact copy is easy, in fact FLAC can store  
an MD5 hash right in the file and can check it when uncompressing to  
ensure the recreated samples are the same as the original. Failing  
that md5sum will do the same thing.

Sure applying DSP to the samples modifies them (Thats what it's there  

  Now to be fair, there are a lot of really nasty CD players out  
there (often sold for 'audiophile' use, that fail to sufficiently  
buffer things to isolate the DAC clock from the raw disk data rate.  
You would think that the way to do a CD player in this age of cheap  
ram was to have the DAC clocked at as close to a constant rate as can  
be managed and then servo the disk read speed to keep a fifo mostly  
full. I have seen designs that did it the other way around!

Shipping CDDA as a master has always struck me as being fatally  
flawed as the format is just not robust, far better to ship a file  
rather then an audio disk. I don't know if there is a standard file  
format for describing exactly what pits go where on an audio disk,  
but it there isn't there should be....

> I was completely serious and sincere about everything I said, but I  
> can see
> that this topic is a non-starter here.

Actually arguing for "generational loss" in a digital system is a non  
starter here, because you didn't qualify that statement. Jitter at  
the AD and DA clocks is well understood, but is ONLY an issue going  
to or from analogue, all the other effects described in that paper  
are down to bit errors during copying which can easily be detected,  
or DSP doing what it does (with varying quality of results).

> I was tryng to bring about a constructive discussion concerning  
> digital
> fidelity and how best to improve it - but it seems like it has  
> elicited
> more of a 'hold-the-fort' response instead.
> That's too bad, because I am just interested in making good music  
> sound as
> good as it possibly can, and demonstrating that independently produced
> music - with OSS tools to whit - can equal or rival commercially  
> produced
> music with their multi-million (billion?) dollar recording studios and
> engineering departments.

Better plugins would help, silly things like lowpass filtering the  
control signals in compressors so as to reduce aliasing in the gain  
control stage,,,,

The real win however is better rooms, instruments, mikes and TALENT.  
Seriously there are recordings that I will quite happily listen to  
that have probably less then 50db SNR and are gone by 8Khz, any  
modern recording chain can do better then that by a huge amount but  
if the talent on either side of the mic doesn't know what to do with  
itself then you will not get a result.

The tools (with the possible exception of mikes, speakers and rooms.  
have not IMHO been the limiting factor for a long time.

Regards,  Dan.

More information about the Linux-audio-user mailing list