- Hardly anyone I know uses the term "dither" to refer to up/down sample conversion.
Dither is a term used to describe something else, having to do with quantizations 
of a bit step. If a bit value is not quite a one or zero, which way does the bit go?
Dither noise equal to 1/2 LSB (lowest significant bit) takes care of that.
That was a point made during a digital audio conference last summer in London
about the use of the term dither. But most of know what you mean by it anyway.
Just a FYI.

- Try to evaluate your coverter somehow. Going 88.2 divi by 2, to 44.1;
just because it's simpler doesn't necessarily translate to better.

- Going 24-bit/96-kHz (pro and mininal for HiRes) to 16-bit/44.1-kHz is fine
and good with the right converter. Shop around, check out some specs, 
find the plug-in best suited for what you're doing (if you're going the plug-in route).


- Partev


--------------------------------------------------------------------------------------------------------------------


--- mle+la@mega-nerd.com wrote:

From: Erik de Castro Lopo 
To: csound@lists.bath.ac.uk
Subject: [Csnd] Re: Re: request for advice: mastering for CD
Date: Sun, 27 Apr 2008 09:28:44 +1000

Denis Crowdy wrote:

> On Thu, Apr 24, 2008 at 11:22 PM, Dave Seidel  wrote:
> > Thanks, Mike.
> >
> >  As I mentioned, I normally render to 96/24, and that's what I play in a
> > concert situation.  I'm guessing that you recommend 88.2K rather than 96K
> > because dithering it down to 441.1K is simpler (just a divide by 2),
> > correct?
> 
> I'm pretty sure the answer to this is no - it is by no means a simple
> divide by 2.  The author of libsamplerate (Erik de Castro Lopo) gives
> a series of lectures to my audio engineering students

I really enjoyed that Denis. Thanks.

> and demonstrates
> how important the filtering and conversion process is so that aliasing
> artefacts don't appear after the conversion,

To expand on this a bit, there are two basic categories of doing sample
rate conversion

  - algorithmically naive and computationally cheap ways
  - good algorithms

When using the first method, yes, downsampling by 2 is better/easier
than downsampling by 960/441.

However, when using good algorithms (ie the SRC_SINC_* in libsamplerate
and similar converters in other libraires and applications) the
difference in quality between downsampling by 2 over downsampling by
960/441 is unlikely to be measurable.

> (similar to those at http://src.infinitewave.ca/) to demonstrate this.

Excellent web site. For a little more on what these spectral plots
mean and how they are obtained, see my blog post:

   http://www.mega-nerd.com/erikd/Blog/CodeHacking/SecretRabbitCode/src_compare.html

A later post

   http://www.mega-nerd.com/erikd/Blog/CodeHacking/SecretRabbitCode/progress.html
   http://www.mega-nerd.com/erikd/Blog/CodeHacking/SecretRabbitCode/rel_0_1_3.html

describes a reworking of the converters I did just recently which put
the best of the libsamplerate converters (with a SNR of 145dB)
extremely close to the best of the commercial converters (160dB SNR or
better). Personally, I think 145dB SNR should be more than sufficient
for any audio application.

Cheers,
Erik
-- 
-----------------------------------------------------------------
Erik de Castro Lopo
-----------------------------------------------------------------
With 22,100,000 legitimate businesses in the US alone,
allowing each to send only one UCE per *year* gets every
mailbox 60,547 emails per day. There will either be email
without UCE or there will be no email.


Send bugs reports to this list.
To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"




_____________________________________________________________
Netscape.  Just the Net You Need.