[linux-audio-user] Re: Companies Refusing to Release/Permit Linux Drivers

[philicorda]

> 
> Message: 3
> Date: Sun, 26 Feb 2006 20:46:50 -0500
> From: plutek <plutek at infinity.net>
> Subject: Re: [linux-audio-user] Re: Companies Refusing to
> 	Release/Permit Linux	Drivers
> To: A list for linux audio users <linux-audio-user at music.columbia.edu>
> Message-ID: <44025A0A.9000104 at infinity.net>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> 
> philicorda wrote:
> 
> >Paradoxically, the only way to avoid digital artifacts is by the use of
> >dithering. This can be proved. There is such thing as correct dithering.
> >  
> >
> perhaps we should fork this discussion off at some point...... anyway, i 
> agree with your assessment of the state of digital fidelity and also the 
> importance of good dithering. i have two questions:
> 
> 1. how would you characterize correct dithering?

1/2 lsb noise. Enough to mask the quantisation noise of digital
recording. This is not optional, it's one of the fundamental
requirements of reproducing a continuous signal with a quantised system.

Fons Adriaensen put it better than I can:

"Yes. A correctly dithered signal converted back to analog is
mathematically
equivalent to the original unquantised version plus some noise. There is
no
way, not even in theory, to detect it was ever quantised. Since it can't
be detected, you can't hear it. But you could fool yourself into
thinking
you can, as many have done before you. After (correct) dithering the
only
'defect' that remains is noise. And with 24 bits and standard signal
levels
this is well below the thermal noise of any analog amplifier that
exists,
and also well below human hearing thresholds."


> 2. are there existing LINUX tools to do it properly?
> 
> back when i used windows, i found the FIR-based noise-shaping dither in 
> "Resample" from SoundsLogical to be much more pleasing to the ear in 
> ambient tails, than other types i had used. unfortunately, the best i 
> have seen in linux-based tools is triangular dither.

Just plain additive gaussian noise or triangular dither does it for me.
I've tried setting up pathological cases to make quantisation artifacts
more obvious, and noise shaped dithers (uv22, waves IDR etc) sounded
worse, though with less overall hiss and slightly more 'depth' to the
sound. I didn't like the way the hiss is all at one end of the spectrum,
and seems to be correlated in some way with the audio, it seems to throw
the recording all out of balance. I found it impossible to tell the
difference when listening to normal music though.

This is totally subjective, and you are probably more critical, so
please investigate this further.


> 
> .pltk.