Syntrillium Backup Forums :: View topic - Dithering to 24-bit?

beetle · Location: USA

If I reduce a 32-bit (16.8 float)file to 24 bit (24-bit packed int), should I be using dither, even if the file will be dithered down further to 16-bit? Is it audiable?

Also, why does CEP have to read 24-bit files as "wav" data when opening?

Syntrillium M.D. · Location: USA

Hey Beetle. Hmm, well, regarding the dither question, I suppose you'd probably want to audition the process on a few tracks and see if there are any accentuated/unwanted artifacts after the final dither down. I would imagine that dithering to 24-bit (so that it can be opened in applications that only see 24-bit files) and then RE-dithering may or may not be a desirable effect and thus may produce undesirable, unnatural dithered artifacts. I can't say that I've had to do that (dither twice). Give it a shot and let me know.

As far as why CEP has to read the 24-bit files as .wav data, i think you are referring to the fact that CEP runs native in 32-bit (16.8 float) and 16-bit, and for those file types creates the .pk file which causes the quick load up of the file. When you are using non-native formats such as 24-bit, the progress window shows that it is opening as wav. Typically, if the file is not an actual Windows PCM wave, Cool Edit will open the file as PCM Raw Data (which is essentially a .wav without the header info)

Hope this answered your question.

---Syntrillium Support

DanT · Posted - Fri Feb 02, 2001 10:22 am

My experience with writing dithering software comes from image processing, not audio, so maybe this response won't be accurate, but here's some food for thought.

Dithering is used when you are processing data in a reduction mode. You may be reducing the number of samples, reducing the precision of the samples, reducing the relative magnitude of the samples, or a combination of the above.

The idea of dithering is to adjust the final value based on the values surrounding the one you are currently working on. Exactly what the dithering algorithm CEP is using, I don't know. But typically you use the nearest samples and weight their contribution to the final result in some way depending both on how distant they are from the sample in question and on how much you want to allow them to affect the sample.

The idea is that you are looking at the context of the sample when determining the reduced value, not limiting your view to the isolated sample - which usually involved a simple truncation or rounding operation.

So, your two-step dithering process should provide you with a better result. That is, you are probably going to stay as close to true by dithering at both reductions than having one of the conversions be a straight truncation.

Now, a caviat. This is highly dependent on the actual algorithm being employed in the dithering process. I don't know if Synt is willing to let their engineers explain to you exactly what that algorithm is - it may be proprietary. In that case, I have to echo Support's reply - try it and let us know. Do the two step dither, and do a straight dither from 16.8 to 16 and see if there are any noticable differences.

I'm assuming you are going to 24-bit to work on it somewhere else and then bringing it back to CEP such that you can't get to 16-bit from 16.8 float directly.

Dan

younglove · Posted - Sat Feb 03, 2001 7:07 am

> If I reduce a 32-bit (16.8 float)file to 24 bit (24-bit packed int), should I be using dither, even if the file will be dithered down further to 16-bit? Is it audiable?

Yes, you should dither to 24-bit first, then
dither again to 16-bit. The dither noise added at 24-bit is much lower than the dither noise added at 16-bit, so it isn't really "dithering twice". The intermediate
24-bit file will have the usual benefit of dithering, that is, you'll hear the original signal at below 24-bit level. Then when you dither to 16-bit, a lot more dither noise is added so you'll lose the original signal at below 24-bit level. BUT, the reason you should dither to 24-bit first is because it makes the dithering to 16-bit smoother. If you don't dither to 24-bit first, the dither at 16-bit can be disturbed and exhibit arthritis, I mean, artifacts.

So dither to 24-bit first to make the 16-bit dither noise smoother, not because you can preserve the sub-24-bit signal at 16-bit (you can't).