SRC and Dithering settings (in general)*

Hi Metoo and Grant,

I'm not sure of what you're hearing or of the finer particulars in how each sample was made but consider this:

Remember how listening to an undecoded Dolby cassette long enough, then turning on the decoding made the cassette sound "dull"? (Leave aside issues of the flaws in Dolby and the flaws in cassette for a moment. This is an thought experiment.)

In my experience, excellent SRC does sound "duller" than most other SRC but not because the former is dull, it is because the latter is bright.

I think the issue really is, which one sounds more like the unconverted original. If you compare against the original instead of against each other, I believe you might find as I did: compared to the conversion from most SRC, the original sounds "dull".

Now, to invert the adjectives for a more correct assessment: the brighter SRC is a departure from the sound of the original, which does not have that brightness.

Best regards,
Barry,
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Thanks for this further explanation of your point. I agree, that is what I was telling Grant.

BTW, if one compares both 44.1/16 samples at around 0:55 the SoX one seems to show a shallower soundstage (I'm talking about reverb of the cello) at that point, while the iZotope one, seems to have more low mids, sounds deeper, if you get my drift.

But, the SoX SRC is quite close in other respects.

 

I asked Grant above if he could please upload a dithered sample of my 44.1/24 file (my mistake here) and seems to have not.

The fact that there is no 44.1/16 version of the SoX file won't allow for any others interested in comparing both SRCs to do so. So, I have uploaded my copy of the SoX file dithered down to 16 bits (I added and 'a' to the file name so as not to confuse it with the one I uploaded before which is still at 24 bits): http://www.yousendit.com/download/U0d3K3BBTXY1UjVMWEE9PQ

I would love to hear the impressions of others comparing these files.

 

Hi Metoo,

I was dabbing a bit with Wave Editor here and can attach your downsampled 24-bit file, now dithered to 16-bit with MBIT+. I used the default setting as shown below. (Linky)

 

Good idea, now others can hear this file at 16 bits with a good dithering algorithm. Just one thing, IIRC you should have used 'Ultra' noise shaping instead of 'medium.'

Again, thanks.

 

Hi Metoo and Antares,

Agreed. With most music, I find the "Ultra" setting, with "Normal" for the amount to work best. You might want to try it and see what you think.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

No problem, thanks for confirming the preferred dither setting Metoo and Barry. Now I guess the files are too scattered meanwhile to expect anyone else to give them a listen. I'll be happy to make another iZotope SRC & MBIT+ Ultra version (and put it together with the SoX & MBIT+ Ultra one in a single upload) if I know which of the options below you guys would like to see used (default shown).

 

Is there a steep linear phase preset? If so, I'd go with that, unless Barry suggests something else. After all, he's the one who has worked extensively with this SRC, I am just thinking in generic terms.

BTW, what program are you using?

 

Sorry I forgot to expand the presets in my screen capture. It's the Wave Editor trial for Mac, which I learned about on this Forum. Sounds logical to use the "Steep, linear phase" preset if a similar setting was used for the SoX comparison file, but I'm not too familiar with the theory behind all this. If nothing else comes up in between, I'll try to post these two wav files in one download tomorrow.

 

Or you could just slide the "Quality" slider all the way to the highest setting.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Yeah, I missed this one. In the SoX settings we have preset VHQ = Very High Quality, so it makes complete sense to move your slider to the highest setting.

 

Okay, here are the two downsampled 44.1 kHz/16-bit wav files for SRC comparison between SoX and iZotope. Metoo's original 192 kHz/24-bit laserdrop is still available from here.

I started fresh from the 192/24 laserdrop in Mac OS X and applied SoX resampling via command line with -v -s -L parameters (very high quality, steep filter, linear phase, no aliasing allowed). The resulting 44.1/24 wav was dithered to 16-bit in Wave Editor, using the MBIT+ Ultra setting to arrive at Nick_Drake-SoX.wav.

The second sample Nick_Drake-iZotope.wav was made from the laserdrop completely in Wave Editor, first applying the iZotope Resampler with the slider all the way up to High Quality (this coincides with the 'Ultra-steep, linear phase' preset) and then dithering to 16-bit with the same MBIT+ Ultra setting.

So, now it's up to more experienced ears with more revealing systems to find the differences. Have fun!

 

Thank you very much, Antares. I hope your work is useful to other forum members so that they can determine directly what algorithm combination would work best for them and why.

 

My pleasure! I really just wanted to play with OS X a bit, which is a big black hole in my computer experience. First impression? It reminds me of a pre-chewed Linux . The bonus is Wave Repair, obviously.

P.S. If anyone wants to download these and the sendspace link has expired, just ask and I'll try to upload again.

 

BTW, Antares, what was your impression of both downsampled and dithered files?

 

As indicated, I probably had more fun making these than comparing the results, Metoo. I'm sure they're both pretty good for what they are and I'm likely to be prejudiced, knowing the settings they were made with. What struck me most was the waveforms, where you can clearly see the loss of detail going from 192/24 to 44.1/16.

If I have to give an impression, the hi-rez file definitely sounds the most realistic. Between the two redbook files, it feels like the iZotope is slightly harder and more closed in, while the SoX is a bit softer and roomier. Maybe I associate this with the "Ultra-steep" filter used for the iZotope downsample. The filter steepness in Sox can be increased from 99% (standard very high quality setting) to 99.7% as well, but the manual warns for a possible increase in artifacts (echo/ringing) this way. Maybe I should have stuck with the "Steep, linear phase" preset for iZotope.

All with a disclaimer of non-trained ears and run-of-the-mill hardware.

 

If you are into it, try the following:
Take the original file and downsample it to 96/24 with iZotope using a 'no aliasing' setting.

Then, take the 96/24 file and downsample it to 44.1/24 using steep 'linear phase.'

Finally, dither down to 16 bits with MBIT+ and the settings mentioned above.

If you do this, compare the resulting file with the one done with SoX and report back.

 

Hi Metoo,

!?

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Yes, I know this looks counterintuitive. You might argue that the iZotope software can take care of the process in one single move (well, actually two, if we factor in dithering) instead of three.

My logic for this is as follows and I would only apply it on a case such as this where the 192/24 original comes from a laserdrop of an SACD. Give that SACDs have a mounting amount of high frequency noise that really starts to get nasty somewhere above 40 0r 45 kHz my idea is to downsample using the antialiasing so as not to end up with distortion from the high frequency noise on the 96/24, or at least as little as possible.

In fact, downsampling to 96/24 tends to do away with most, if not all, of the visible high frequency noise seen on the frequency spectrum graph. My guess is that 88.2 would totally eliminate it in all instances. This is probably why all HiRez players that convert DSD to PCM on playback do so at 88.2/24.

Then, with a clean, or at least cleaner file you can do a normal downsample to 44.1/24 and then dither to 44.1/16. You'll probably ask, why not just downsample with no aliasing directly to 44.1 kHz? Apart from the reason mentioned above, even though I know that the quantizing filter should take care of the frequencies above the set frequency point, I usually aim at 96/24, not 44.1/16 with most of my SACD laserdrops.

My suggestion to Antares is because I am surprised that he seems to notice more quality in the SoX algorithm than on the iZotope one (I hear the opposite of what he describes on each case). Could it be due to aliasing? Then again, he has gone the command line way, instead of the Foobar plugin, although his settings seem to be quite in line with that on the plugin.

 

Hi Metoo,

Well, I'd never say not to try it but personally, I see/hear it differently.

You are assuming in the first step that SACD noise is gone because you don't see it in the software you are using to view the frequency response. Two things there: first the software and second, you are forgetting the noise could intermodulate down in the audible spectrum - so the fact that you don't see a rise with your software should not, in my view, be taken to mean the noise is "gone".

It is probably impossible to determine why one person hears a comparison one way and another hears it the opposite way. The DAC, its connection to the monitoring gear, the monitoring gear itself, how it is set up, the person's hearing sensibilities and listening biases all come into play. I would not necessarily ascribe the different perspectives to aliasing. In my view, there are too many other factors that will be much more prominent. Especially if you listened to the same exact files and reached a different conclusion.

One of the main principles I've always used with audio engineering and the one I emphasize to my assistant and folks I speak with all the time is:
"For everything you do, every process, every decision, first have a good reason for doing it or deciding it. In the absence of a really good reason, don't do it. Ask yourself why you would perform the process and what you might have done wrong in an earlier step that would lead you to where you are now. If you can think of something, go back to the earlier step and re-do what was done wrong the first time."

Just my perspective.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Hi Barry,

I was/am aware that the fact the noise is not that apparent in the graph does not mean that it is totally gone. I simplified my reply so as not to write a book.

My point is that I feel that I can reduce the intermodulation that can happen as a result of downsampling a file with a lot of high frequency noise by using antialias and its non-steep curve. This should eliminate more of the unwanted high frequency noise than would be the case if I went with a steep downsampling that does not take into aliasing into consideration.

Notice that I leave the steep downsampling for the next step, so as to keep as much high frequency information on the final 44.1/16 file.

If you see/hear this differently I would sincerely like to know how you would approach this case. But please center your reply on methods/approaches I can apply.

I am aware of the importance of the software and hardware used. While, I, and you, and everyone else that cares for quality results would love to use the best tools available at any given time, unfortunately I just have to make do with the best software and hardware that I can have access to at this moment. So, I am striving for the best results with what I have, which at the end of the day is what it's all about.

Good, sound, advice.

 

Hi Metoo,

I'm not worried about intermodulation during SRC. I'm saying the damage has already been done on the original.

In keeping with my saying about having a good reason for every process and decision, I believe the additional application of SRC you propose would not provide any benefits and would subject the file to a needless additional process.

Applying another steep filter won't undo the damage. It will spread the information out in time, adding a bit of a "smear".

That said, please don't let me stop you. As I said in my earlier post, I'd never say to not try something. It just isn't the path I'd personally take.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

Let's see if I understand your point correctly. You are saying that the intermodulation problem is already present in the DSD original? If so, I imagine that you mean that even though the high frequency noise has been noise shaped it is still affecting the frequencies we hear, right?

If this is the case, we can conclude that all the DSD-derived masterings on CD by Sony are 'compromised,' as are all the DSD-sourced discs.

In my case, when I do a laserdrop of an SACD what I want is to get as close to the sound coming out of my player as possible (given that I cannot access the original data stored on the disc) and convert that to a - to me - more flexible format: PCM. Since I have not done the original, I will have to 'put up' with the limitations of the source and my current playback chain.

Evidently, and as I have mentioned on this forum before, any serious/committed recording exercise that looks to get as close to the sound of the source as possible, be it an SACD, a vinyl LP, or your own recording, is a veritable learning experience. I am open to learn and, as always, will be able to better comprehend those parts of the experience that I can carry out myself or hear/detect.

Evidently, you are mostly working with recordings you did directly from the real source. Most people here are doing needledrops, laserdrops, and cassette recordings or backing up old tapes of their musical projects (as is also my case). In all these latter situations we are normally working with someone else's recording and in a predetermined format (even when I record from my analog multitracks). This is a ways away from your direct experience with live musicians in an environment you decided on, and with the miking of your choice.

In this sense, your input is invaluable. Given you have had access to the live source you have a better grasp of how things should sound and can more easily determine if a certain software and/or hardware is delivering the goods.

The fact that our only reference is a pre-existing recording by definition introduces variables we cannot control. Thus, our sound quality reference is how close our recordings sound to the sources we have used. It is also the essential experiential gap we cannot close between our usual audio experiences and yours, Yet, it is the limited 'laboratory' from where our decisions/conclusions and methods come.

 

Hi Metoo,

I wouldn't tell anyone what to conclude. The conclusions I reach are only for myself (and my clients). To my ears, SACD is a compromised format. Better sounding than CD in many ways but for me, it falls far short of what the best PCM can achieve. (I also differentiate between DSD and its distribution medium SACD. Though I'm not a fan of either one.)

I too would want to get as close to the source's sound as possible. This is why, as I mentioned in an earlier post, I wouldn't introduce an extra generation of filtering.

This is not true. When I'm mastering a recording someone else made, I have finished mixes to work with -- someone else's recording and mix, often in a "predetermined format", the product of their hardware and software (or in the case of analog, hardware alone). When I'm mixing, I have tracks that are already set in their "predetermined format".

In those cases where the recording/mix is very good, I want to preserve it as is. That means I'm going to do a little as possible in the mastering process. If I need to make level adjustments and the source is not high res, I'll increase word length, then up the sample rate to make the adjustments. On the way back down to CD, I will only convert once. Even the best gear, good as it is, is not perfect.

Just my perspective.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

OK, I understand. I will test doing it in one step with no aliasing.

You are right. Somehow when thinking about the whole enchilada I missed that 'small' part of your work. I stand corrected.

Thanks for your reply, Barry.