24bit to 16bit, what is the best way to do it ?!

[Acix]

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif ...that's what I use. .

Yep...Gregorio need Wavelab too.

 

[gregorio]

Quote:

Originally Posted by panda /img/forum/go_quote.gif i am interested in what the sound editing gurus suggest for resampling files. i need to make 24/96 into 16/48... would appreciate advice on achieving the best way. to use foobar via resampler plug in and 16bit dithering is the most simple, just wondering if there is a better sounding method. sound forge, adobe audition, etc???

To be honest, I am not that well versed in consumer or pro-sumer programs like Foobar or Wavelab. I would suggest you convert from 24/96 to 16/48 and if possible use a noise-shaped dither. Provided you are not going to be doing any additional processing, such as EQ, I would recommend using the strongest noise-shaping algorithm available. On profession dither programs you usually have the option of Type 1, type 2 and type 3 plus the option of switching off the noise-shaping. I would recommend the type 1, this will yeild the most agressive redistribution of the noise to outside the the critical frequency bands.

I notice that wavelab uses UV22, which was very highly regarded professionally about 15 years ago. I don't hear of it so often now though, so perhaps they haven't kept pace with developments. If you can find some software that includes PowR dither, that is a ninth-order noise-shaping dither which is not the very best but still good professional quality. The best professional ones, tend to be available as plug in processors (for ProTools for example) and are likely to set you back $500 or so.

G

 

[leeperry]

actually ffdshow has 4 levels of "floating point to integer" dithering(I EQ my stuff to kill ear resonances so it goes 32float>24integer) :
-no noise shaping
-low noise shaping
-medium noise shaping
-heavy noise shaping

I think I prefer "low", but is there any audio test pattern to compare them

 

[gregorio]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif actually ffdshow has 4 levels of "floating point to integer" dithering(I EQ my stuff to kill ear resonances so it goes 32float>24integer) : -no noise shaping -low noise shaping -medium noise shaping -heavy noise shaping I think I prefer "low", but is there any audio test pattern to compare them

Going from 32float to 24bit integer is hardly going to need noise-shaping as all the quantisation noise should be below -144dB, which is way below the noise floor of any component in your playback chain. Unfortunately, it's a little difficult to know exactly what goes on with 32bit float. It's not always a simple conversion because of how the floating point of each channel has to be rounded to fixed integer. High quality professional equipment always tends to be fixed integer.

Going from 32 to 16bit though would probably benefit from heavy noise-shaping but you'd have to have your amp very high to notice.

G

 

[leeperry]

well apparently each float<>integer conversion is VERY lossy...how do you compare noise shaping algorithms anyway? coz it'd take a lot of practice to hear any difference, from my tests in Wavelab & ffdshow

 

[gregorio]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif well apparently each float<>integer conversion is VERY lossy...how do you compare noise shaping algorithms anyway? coz it'd take a lot of practice to hear any difference, from my tests in Wavelab & ffdshow

Well, it absolutely shouldn't be that lossy! With 32bit float and 24bit we are talking about the most minute of levels. -144dB (24bit) is about the same level of sound created as the noise made by the electrons colliding inside a 1.8k resistor!! It is way beyond the ability of any playback device on the planet. Something dodgy is going on somewhere in the process.

You cannot directly compare the difference between noise shaping algorithms as they can't be heard! The easiest way is to take a 24bit recording and reduce the gain by about 90dB, then put it through a noise-shaped dither algorithm into 16bit. Lastly, play the 16bit file with your amp turned up by about 90dB. This doesn't give a real world impression of what the dither will sound like but it does allow you to compare and measure differences and identify any artefacts.

G

 

[leeperry]

Lahey - Floating point

Quote:

Conversions to integer can unmask inaccuracies in a floating-point number, as is demonstrated by the next example.

oh ok, so comparing noise shaping algorithms is next to impossible...that's good, so I got no idea which one to choose in ffdshow

 

[gregorio]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif oh ok, so comparing noise shaping algorithms is next to impossible...that's good, so I got no idea which one to choose in ffdshow

The real difficulty with most professional dither algorithms is that they are normally implemented as part of a mastering limiter (Waves, Sony, Maxim, etc). Limiters cause colouration of the sound so it's very difficult (read; next to impossible) to specifically compare the dither algorithms.

As what you are doing is just for your own use, I would take a very quiet section of one of your 24bit tracks, copy it a number of times, then apply a different dither algorithm to each copy and then turn up your amp and compare them.

G

 

[Acix]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif The real difficulty with most professional dither algorithms is that they are normally implemented as part of a mastering limiter (Waves, Sony, Maxim, etc). Limiters cause colouration of the sound so it's very difficult (read; next to impossible) to specifically compare the dither algorithms. G

I use Apogee UV22HR with great results. And if there is Something better out there, I will be happy to test and compare...
http://www.apogeedigital.com/pdf/UV22osquick.pdf

About the L2 L3...they are nice do the job, but too much aggressive for me.

 

[gregorio]

Quote:

Originally Posted by Acix /img/forum/go_quote.gif I use Apogee UV22HR with great results. And if there is Something better out there, I will be happy to test and compare... About the L2 L3...they are nice do the job, but too much aggressive for me.

I used the L2 for quite a few years, it can't be too agressive as it's completely adjustable. I don't think I've heard the UV22HR version but my guess is that at normal listening levels it's indistiguishable from any other professional dither. There must be some other factor at work, levels for example but there are a number of other potential factors.

The world has changed, just a few years ago consumers had no access to hardly any of the tools we used professionally. With all due respect to you and others, I'm not sure that putting professional tools in the hands of consumers is such a great idea. They will generally not have the experience or technical knowledge to apply them to best effect. At the end of the day, all these processes and processors are just tools and as such only provide their expected performance in the hands of a craftsman. I'm not trying to be condescending, just realistic as regardes the vast majority.

G

 

[grawk]

That's pretty freaking elitist.

 

[Arjisme]

Is it? Sounded like a reasonable generalization to me.

 

[JaZZ]

Quote:

Originally Posted by grawk /img/forum/go_quote.gif That's pretty freaking elitist.

I agree!


Quote:

Originally Posted by Arjisme /img/forum/go_quote.gif Is it? Sounded like a reasonable generalization to me.

The same as «computers are something for computer scientists» -- in the eyes of a computer scientist?
.

 

[Aleatoris]

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif I agree! The same as «computers are something for computer scientists» -- in the eyes of a computer scientist? .

Nah, more like AI programming is for a computer scientist, in the eyes of a computer scientist. As a note, most of the comp sci students that I know... Know nothing about computer hardware.

 

[Arjisme]

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif The same as «computers are something for computer scientists» -- in the eyes of a computer scientist? .

More like how your average consumer who buys a pro-level camera still isn't likely to create pictures appreciably better than what they could with a decent consumer-based camera.

As I said, it was a fair generalization which, of course, allows for exceptions. There is plenty of room there for the preservation of egos.