Digital Artifacts when downsampling to 8k

[Vamalla]

I'm not sure if I'm the only one this is happening to but when I downsample to 8k, my .wav's are constantly filled with artifact-static when peaks rise (not clipping), when somebody took the same file and downsampled in SoX, the .wav didn't have the same "artifact" qualities. Then had QA show me, and insert face-palm here. Made sure that "Quality" was at 100% too. The last thing I want to do is ditch Audition, but if this keeps happening, I'll have to!
I'm downsampling from 48k to 16k for telephone prompts (updating system to 16k!) I still need to maintain the older systems (8k) and whenever I downsample to 8k from 48k, I get this problem. As I'm sure you can guess, I don't have that problem when going to 16k.

Any advice will help

[SteveG_AudioMasters_]

The moment you downsample anything, you're going to get more artifacts - that's inevitable. But if you get a different result when going to the same format using a different approach, then I'm a bit suspicious, I'm afraid, as that simply shouldn't happen.

The first thing we need to know is exactly what formats you are converting from and to, and all of the rates concerned. Even within the same apparent format, you get options sometimes. And, we need to know what you've started with, in terms of sample rate and bit depth.

The second thing is to ask you what you've done to the original files to optimise them for low bit rate saving, as what you do to them will make a significant difference to the results. The basics of it are that you have to reduce the dynamic range considerably - which generally means no peaks of any significance, and that the overall level is kept as constant as possible.The idea is that you get all of the wanted signal into 6 or 7 bits as far as the amplitude is concerned, because that's all you will get when you downsample. And yes, you do all this to the original, not to your downsampled version. I know it sounds a bit contra-intuitive, but believe me, you want to start with the best original you can when you save in a restricted format, so you don't have to mess about with it at all after you've converted it.

With this sort of situation, you really can't give us too much information!

[Vamalla]

So I found out that the file in question was running @ 16bits and not 8, so everything got thrown out the window

Under "Sample Rate Conversion" there's a drop-down called "Advanced". What does this do? It doesn't seem to have an audible difference to me, I could be wrong. And what does it mean to Pre/Post Filter? I'm trying to preserve quality throughout Conversion for obvious reasons. Would dithering help?

I'm sorry for throwing a ton of questions but my goal is to be absolutely proficient in resampling through Audition. I'm working mainly w/ IVR systems. I'll downsample to 8k from 48k but I'm very cautious about losing "extra" things that I don't need to lose. Thank you for the help Steve, I knew something was out of place

[SteveG_AudioMasters_ · Accepted solution]

Vamalla  wrote

Under "Sample Rate Conversion" there's a drop-down called "Advanced". What does this do? It doesn't seem to have an audible difference to me, I could be wrong. And what does it mean to Pre/Post Filter? I'm trying to preserve quality throughout Conversion for obvious reasons. Would dithering help?

The one thing that might well help is dithering, especially with a file that's going to end up as 8-bit. You have to be careful only to do this once though, so you leave it until you create the final 8-bit file, and don't get tempted to do any more to it afterwards. What dithering does is to artificially make the transition to the least significant bit appear to be a lot 'smoother', appearing to extend the apparent bit depth.

AS for the filtering settings, leave them as they are - this is all to do with sums when it gets past 50%, and pretty much irrelevant when it comes to an 8-bit conversion anyway.

But you really do have to make sure that you've done everything before the final conversion, so that when you lose those bottom eight bits, it really doesn't make any difference. And to do this you have to have a reduced dynamic range signal, so that nothing falls below about -40dB (or even higher if you can) on your original, or you'll just end up with the horrible noise-floor noise obscuring your audio. Generally speaking, once you've realised what it is you have to do, it becomes almost second nature to process the spoken word this way.