Topic: Convert to 32-bit?

1. Convert to 32-bit?

In the GPO manual, it mentions in the section on the options menu in Kontakt that there is a switch to convert samples to 32 bit, but it doesn't explain the advantages/disadvantages of doing it. I assume it is higher resolution. What is the resolution of the original file? Is there a way to premanently convert the samples to 32 bit, so it doesn't have to be done every time and the load times are faster?

Sorry for all the newbie questions, but a search of the forum didn't produce an answer.

Thanks.

2. Re: Convert to 32-bit?

I hope we can get a good answer to this. I was just getting ready to post the same question. It seems like the answer could depend on the format you're recoding, for example I usually record at 48K/24bit, but that could mean that the samples get converted once to 32bit and then again to 24 bit. I need to understand this as well. I'm pretty sure the original samples are 16 bit and at first thought the conversion from 16 to 32 bit seems pretty non-destructive, but I'm not sure that's the case. Any help understanding this issue would be greatly apreciated.

3. Re: Convert to 32-bit?

The conversion to 32 bit from 16 bit or 24 bit samples is straightforward to say the least. The input data is simply padded with extra bits (16 in the case of a 16 bit sample, 8 in the case of a 24 bit sample). Note that this doesn't add any extra resolution or quality to the original sample material, but it may take up more memory than native 16 bit samples. The advantage to this lies in how Kontakt (or any other DSP application) manipulates the samples internally. When Kontakt does some operation on loaded samples (like increasing amplitude or applying reverb effects), the number of bits the samples have determine the precision of the result. More bits = better accuracy of the new sample data. So mathematically speaking it's a good idea to use 32 bits, but the real question is whether or not you could actually hear the difference. Using 32 bit resolution is not a free lunch. The samples take longer to load, and (depending on how Kontakt implemented their effects engine) could take longer to process.

The best thing to do is try and load the same set of samples as 16 bit and as 32 bit, play whatever it is you're playing, and see if you can hear a difference. Pretty much everyone here says to turn the thing off, so I'm guessing there's not much difference in the sound.

4. Re: Convert to 32-bit?

Easy_Rhino, Thanks for the answer. Clear and informative.

5. Re: Convert to 32-bit?

Originally Posted by Easy_Rhino
The conversion to 32 bit from 16 bit or 24 bit samples is straightforward to say the least. The input data is simply padded with extra bits (16 in the case of a 16 bit sample, 8 in the case of a 24 bit sample). Note that this doesn't add any extra resolution or quality to the original sample material, but it may take up more memory than native 16 bit samples. The advantage to this lies in how Kontakt (or any other DSP application) manipulates the samples internally. When Kontakt does some operation on loaded samples (like increasing amplitude or applying reverb effects), the number of bits the samples have determine the precision of the result. More bits = better accuracy of the new sample data. So mathematically speaking it's a good idea to use 32 bits, but the real question is whether or not you could actually hear the difference. Using 32 bit resolution is not a free lunch. The samples take longer to load, and (depending on how Kontakt implemented their effects engine) could take longer to process.

The best thing to do is try and load the same set of samples as 16 bit and as 32 bit, play whatever it is you're playing, and see if you can hear a difference. Pretty much everyone here says to turn the thing off, so I'm guessing there's not much difference in the sound.

Here's my understanding of it... the setting should not affect the output sound quality at all, because the sound is always processed as 32-bit data anyway. The question is WHEN the samples are converted to 32-bits.

Stored on disk as 16-bit samples, but then padded to 32 bit when loaded into memory, would double the amount of memory required to store the samples.

However, if the samples are loaded into memory as 16-bit data, it's more compact... but the CPU has to do more work each time it retrieves the samples from memory for playback. Each 32-bit memory address holds two 16-bit samples, so it has to go to the correct memory address, break it into two samples, pad both of them, and then they are ready for processing by the software.

If the samples are padded as they are loaded into memory (doubling the memory requirement) it saves the CPU from having to do this extra work during playback.

So it's basically a tradeoff between memory usage and CPU utilization. I suppose I'll play with it later on and monitor the memory and cpu usage.

6. Re: Convert to 32-bit?

Could this possibly be an answer to people's "snap, crackle, and pop" problems?? If we convert the samples IN MEMORY, will that significantly reduce the amount of processor power needed upon playback?

If so, that might be excellent for people with a lot of RAM but clicking/popping due to processor overload.

Can't wait to experiment when I get home...if someone KNOWS, please reply!!

Jim

7. Re: Convert to 32-bit?

Originally Posted by snorlax
Could this possibly be an answer to people's "snap, crackle, and pop" problems?? If we convert the samples IN MEMORY, will that significantly reduce the amount of processor power needed upon playback?

If so, that might be excellent for people with a lot of RAM but clicking/popping due to processor overload.

Jim
My processor is constantly getting hammered to the ceiling, in that I use the GPO in conjunction with Finale 2005, doing orchestral scoring with moderately large ensembles.

Has anyone seen any significant improvement in processor loading via this approach? I've tried it, but as I'm running a somewhat older machine (2.2 GHz, 1G RAM), I'm not quite sure whether it's helping performance or hurting it... lol.

8. Re: Convert to 32-bit?

Originally Posted by etLux
My processor is constantly getting hammered to the ceiling, in that I use the GPO in conjunction with Finale 2005, doing orchestral scoring with moderately large ensembles.

Has anyone seen any significant improvement in processor loading via this approach? I've tried it, but as I'm running a somewhat older machine (2.2 GHz, 1G RAM), I'm not quite sure whether it's helping performance or hurting it... lol.
Man, I'm still waiting for this answer as well.
I tried it on a 2.4 gHz P4 and didn't notice much, but let me try again!!
Jim

9. Re: Convert to 32-bit?

I tried an experiment with this setting on and off. It definitely uses more memory, although it was hard to tell if it really was exactly 2 times for 16 bit samples or some lesser value. I really didn't notice much of an effect on processor time (or at least enough of an effect to be detected by the standard Windows CPU monitoring tools), but I swear the output using the 32 bit samples was less, um, "brittle" sounding. It sounded suspiciously like digital aliasing from a rather hamhanded 32 bit to 16 bit conversion. Unlike the 16 to 32 bit conversion, the 32 to 16 bit conversion is way, way more complicated. You see, if you start with 16 bit samples, then pad them with another 16 bits (creating a 32 bit word), and then perform a bunch of arithmetic operations on it, your result will most likely utilize some or all of the extra bits. The problem is now, if you want to convert the result to 16 bits, what do you do with the extra bits of information? The easy solution is to simply chop them off, but then you get a lot of "noise" in the resulting 16 bit audio sample. The more complicated solutions involve algorithms to "dither" the extra information into the audio stream using a random sequence of 1s and 0s in the 16th bit.

So, assuming I wasn't just hearing things that weren't really there (the "wishful thinking" placebo effect), my guess is that if the 32 bit samples flag is checked, all 32 bits of information are passed back to the sequencer (and ultimately the soundcard) by Kontakt, otherwise the result is scaled down to 16 bits by Kontakt before passing it on. For owners of soundcards that are capable of handling 24 bit audio samples, this could result in a better sounding output. If you have a 16 bit soundcard, the soundcard drivers probably take care of it, so there may not be any audible increase in audio quality unless the drivers utilize a better dithering algorithm. But if you keep the audio stream at 32 bits for the whole project, then you have the opportunity to choose the dithering algorithm when you master the result rather than be stuck with the one that Kontakt uses. For example, Ozone 3 uses MBIT+ dithering, which yields about as good as a result as you're going to see.

10. Re: Convert to 32-bit?

Originally Posted by Easy_Rhino
For owners of soundcards that are capable of handling 24 bit audio samples, this could result in a better sounding output.
Easy,
I tried an experiment comparing the output of both the 32- and 16-bit modes and found there is no difference. I have a 24-bit soundcard and did the experiment in 24-bit mode.

I loaded a sample at 32-bit and recorded a wave.
I loaded the same sample at 16-bit and recorded a second wave.
Inverted the phase of one wave and generated a composite wave.

The output of the composite was null, meaning that the samples at 16- and 32-bit do not produce different output.

I did another experiment on load times of the 16- and 32-bit samples, using my Full Benchmark set. The set loaded noticeably faster at 32-bit than at 16-bit, but given the significant extra memory for no noticeable benefit in sound quality, I've stayed with the 16-bit.

Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•