SRC and Dithering settings (in general)*

I wonder if a 64-bit SRC will work correctly in what must be a 32-bit computer. Given that Barry's is 5 years old, does it have a 64-bit processor? If not, and there is a way for a 64-bit program to run on a 32 bit computer I guess that would explain the slowness of the process.

 

Steve, I imagine that we are all using the Foo DSP resampler as the SoX resampling tool with Foobar, right? I ask this, because I suddenly started doubting if there was a way to plug in the original SoX program that. AFAIK, only works through the command line.

I also imagine that said Foobar plugin contains the resampling core of the original program, right?

This said, I am somewhat worried about the 'High Quality' option in the Foobar plugin due to the fact that the original program shows some weird double triangle at the top of the sweep with this setting. It seems that the Linear phase setting, which I have also chosen when setting up the plugin preferences works fine (quite similar to iZotope's) at least on the source program. All this is as per the graphs at the well-know SRC Comparisons website: http://src.infinitewave.ca/

Have you any information regarding these things?

[Note: Sorry, I just found out that the Linear phase graph is really using the Very High Quality setting, which they have mentioned as 'VHQ.' Yet, I still have the question about how much of the original program in in the Foobar plugin.]

 

Grant, have you done a critical A/B comparison between the results of the SoX SRC and the one in Audition?

 

Correct on all counts!

 

Hi Stefan,

Thank you for your kindness.

It is unfortunate that Windows users must spend $1200 to get an "almost as good" version of the SRC (32-bit vs. 64-bit), which is available (along with the excellent MBIT+ dither/noise shaping algorithm) to Mac users for only $79 in Audiofile Engineering's Wave Editor.

WE was on sale this past December for just $59! Even if one didn't use the rest of the program, for the algorithms alone, I deem it a steal. In my experience, the 64-bit SRC and MBIT+ outperform the "expensive pro products" by a country mile. When price is factored in, even if you include a small Mac to run it in the bargain, there is no contest. (I have lots of friends who use Windows for audio but my personal perspective is that it is not really ideal for this purpose -- at least not when compared to OS X or Unix and at this point, there is considerably more software for the former. And all the state of the art hardware -to my ears anyway -is designed for OS X too. Just my experience. I know and respect that other folks feel differently.)

Regarding speed, in a discussion elsewhere that is going on at the same time as this thread, I have found that even on a new Mac, the SRC isn't much faster.

MBIT+ is also available for Windows users in the Ozone plug-in but still it costs four or five times as much as Wave Editor does.

SoX may be a good value but for my ears, while it sounds better than many others (most of which are not good at all), it does not have iZotope's amazing absence of spurious harmonics, which I believe makes the latter able to create files that do not have the added brightening (sometimes hardening) I hear in other SRC algorithms.

I'll be happy to listen to the files once I get a chance but that may be some time in the future as there is a lot on my plate at the moment.

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

 

Hi Metoo,

I believe there is some confusion (perhaps on my part).
The 64-bit SRC is not a 64-bit application. As I understand it, it uses 64-bit math internally.

In the same way, soundBlade, running on a 32-bit processor machine, does its internal math at 48 or 64-bits, depending on the process. And ChannelStrip does its math at 48-bits. And the Console software for Metric Halo interfaces does its math at 80-bits.

This could indeed offer a partial explanation but I have other decent SRC that is 32-bit and is just as slow. And the MIO Console processes are virtually instantaneous. I believe credit goes to the programmer, who is a true genius.

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

 

Hi Barry,

Thanks for your words (to Stefan) above.

Yes, it is a pity that us PC users can't get even a standalone version of the iZotope SRC at a sensible price. I am starting to believe that iZotope may be missing on a large market there by not coming out with low-priced standalone SRC (and perhaps dithering) software for PC.

This said, I understand that business and other more important matters should always come first. I do want to take a moment to thank you for the time you have been giving this thread and the previous downloads.

 

Barry, I might be mistaken, but I believe that regardless of the processing bit depth a program uses if you are on a computer powered by a 32-bit processor it is this chip that must handle all the calculations, and will do so at its 32 bits. So, in a way, it would be like inserting a bottleneck in the processes.

 

Hi Metoo,

I am grateful for the perspectives I get from you and other listeners in this thread (and in this forum in general).

In discussions with the designer of the Metric Halo hardware and software, I have come to understand that it isn't merely an issue of losing sales to the large market represented by Windows users. He tells me he also loses all the support calls and associated headaches, which would be many. These would divert him from the tasks of continuing development and add overhead in terms of support folks. They are a small company and must choose their priorities.

iZotope does license its 64-bit SRC (and MBIT+ too, I believe). Perhaps the thing for Windows users to do is to contact the companies that make their favorite software and ask them to license these. ?
On the Mac side, companies like Audiofile Engineer and even Sonic Studio license the 64-bit SRC (and MBIT+) recognizing them as the wonderful achievements they are and in this way, allowing their customers access to them.

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

 

I have just realized that the 44.1 WAV file I uploaded yesterday is not at 16 bits, but at 24, which means it has not been dithered to anything, my bad.

Grant, would you be so kind to dither yours down to 16 bits and see if you hear a better sound?

I am currently trying to get a guy at another forum to download and resample+dither my 192/24 file using RX Advanced. Will report back if he accepts.

 

OK, he was gracious to help out and here's the link he sent me: http://www.megaupload.com/?d=L9REQ7VR

This is a downsample of my original 192/24 file done with RX Advanced using iZotope's 32-bit SRC and MBIT+ dithering. He tells me, 'I used the "Steep, no alias" SRC preset and the normal dithering setting with MBIT+ Ultra.'

BTW, his uploaded file is also in WAV format.

I have just downloaded it myself and will proceed to compare it with my SoX-downsampled file. I would like to know your impressions after having compared both (or the three) files. Could someone who has downloaded my 44.1/24 and has MBIT+ dithering capability dither my file and upload it so that everyone can hear things as they should? Thanks in advance.

Finally, I'd like to thank the person from the other forum who contributed this file. I think this will let us hear how the SoX SRC stands vis a vis a current industry standard as is iZotope's SRC.

 

Has anybody here compared these files?

 

I just noticed that the iZotope SRC-downsampled file was done with at least two settings that are similar the ones I have preset in the SoX SRC:

- I have the 'Allow aliasing' option unchecked in SoX, which I imagine would be the equivalent of the 'no alias' of the iZotope SRC.

- And I have the 'Steep filter' setting in SoX, which I imagine is equivalent to iZotope's 'steep' setting.

I have read on the SRC Comparisons page that iZotope's SRC has different phase settings and I do not know which one was used in this file, but I have Linear Phase response checked in the SoX preset interface due to what I commented at the start of this thread.

 

They both process at 64 bits internally. However RX only inputs and outputs to a 32-bit float audio format.

Yes, the 64-bit precision of floating point calculations is achievable since 80486 processors 20 years ago.

This does not relate to floating-point calculations, only to the integer math. So, I wouldn't expect any difference in the processing of audio applications.

 

Hello Alexey,

Thanks for your input clearing up these doubts. As always, your posts are very interesting.

Just one question: Why, then, are computers taking so long to downsample with your SRC, as is the case with Barry Diaments'?

Thank you in advance for your reply.

 

On my 3 GHz computer, RX performs SRC about 10x faster than realtime at its highest-quality setting. The only exception is a non-linear phase mode, it can take longer, but still runs faster than realtime here.
Maybe Barry is not using the latest version? In one of the first RX versions, SRC used to be much slower, but it's been optimized since then. Please PM me the details, if you think your processing speed differs significantly, we can try to investigate it.

 

Now that you mention phase modes. From what I have read regarding them I tend to favor linear phase. Are there any advantages/specific uses of minimum and intermediate phase?

Again, thanks in advance for your reply.

As to Barry's current processing times, I hope he finds a quick solution; time is precious especially when working under a deadline. I imagine he will get in contact with you as soon as he drops by this thread again.

 

Hi Alexey,

Thank you for the clarification.

Metoo, I spoke with B.J. Buchalter, designer of the Metric Halo products and he said essentially what Alexey just said.

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

 

Hi Alexey,

I'm using the 64-bit SRC that is part of Audiophile Engineering's Wave Editor (the latest beta).

Are you saying RX is faster? You mention a 3 GHz machine. My PowerBook G4 runs at 1.25 GHz. Could that account for the difference? Or is RX faster anyway?

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

 

Isn't the world just a wonderful place to learn? One never goes to bed without having learnt something new.

Thanks, Barry.

BTW, I hope this info about your processing time with the SRC that Alexey picked up on here is of help to your work.

 

From what i'm hearing, I get about the very same 16-bit, 44.1 results whatever is used for the process. But, I do sense that the one your buddy did sounds a tad veiled. Maybe not, it's just what I sense.

 

Yes, I also heard what could be expressed as 'veiled,' but I believe it might rather be the smootheness of iZotope's SRC as compared to what Barry described above (post # 80) as 'the added brightening' of other not-as-transparent SRCs.

In any case, that their results sound quite close is interesting given the price of SoX (the poor man's 'good' SRC). I also don't believe that this fact diminishes the value of Alexey's great programming at all.

 

Grant, have you noticed the tambourine on the left channel towards the end of the song? That's a cool detail I had never heard until last night.

BTW, do you know how to subtract one file from the other to see/hear the differences between both SRCs? That'd be a good other test to perform.

 

No, well, I don't remember. All I know is that I have some laundry to do, food to eat, and some TV to watch before I go to bed.

Later!

 

Hi Metoo,

If only that was true of our politicians.

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