SRC and dithering

[bdiament]

This is in Hardware but is really about Software.
Not the kind we listen to directly, like a CD or record but the kind used to create the ones we listen to. I'm talking about mastering tools.

I've spent most of today evaluating some new software I have, particularly algorithms for sample rate conversion and dithering.

Both processes come into play when creating a CD master. Since I record in high resolution at 24 bit and 96k sampling (with some current experimentation at 192k as well), in order to create a CD master, sampling rate must be brought down to the CD standard 44.1 kHz and word length must be brought down to the CD standard 16 bits.

Even when mastering an outside project recorded a lower resolution, if any sort of processing is necessary (as it often is with the output of most modern studios and mixrooms), I will create a high sample rate, longer word length version on which to perform all processing. When this is done, again, I have something that requires sample rate conversion and dithering.

What I found today, comparing what I deem to be the best of each type of process (two sample rate conversion algorithms and fourteen dither algorithms) is that among the best available, SRC makes a bigger difference. I should add that when I say an algorithm is "better", I mean it sounds closer to the original source.

Interestingly, what I'm hearing as the best SRC and best dithering -both significant steps up from what have been the best of each for the past several years- both come from the same source. A young Russian designer named Alexey Lukin. His iZotope SRC represents a big advance in artifact free sample rate conversion.

I am still comparing the dither algorithms but have so far found that the three that stood out for me today among the 14 I'm listening to are all variations of Alexey Lukin's work too. I still need to listen to these some more (today has been a long, long day) and will report when I reach a conclusion.

Meanwhile, for those who wish to hear what I was listening to as far as sample rate conversion, visit:
http://www.soundkeeperrecordings.com/zwsrc.htm

There you will find three samples. One, the original 32 bit float, 96k file which I fed toeach of the conversion algorithms. The other two are the 32 bit float, 44.1k outputs from the processes. (I'm not sure if a Mac will be required to hear these properly. Different software does 32 bit differently.)

Best regards to all listeners,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

[LeeS]

Great background and results Barry. Thanks for sharing. :righton:

[thxdave]

Barry, as a fellow MH user, I'm very interested in hearing your results.

[Grant]

Thanks for the thread! As you know, I take this stuff seriously too.

[bdiament]

Can you hear the files (either directly or after download)?

Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

[Metoo]

Although iZotope 64-bit SRC is a Mac product and I have a PC I am totally interested in reading/listening to your tests, results and conclusions. :)

[Metoo]

Barry, I have just finished downloading your first aiff 32-bit file and find that I cannot play it. I tried playing it with Quicktime and all I get is a kind of shaky hiccupy sound. My other programs do not even recognize the format. :(

Note: I could finally listen to it by converting your aiff file to WAV in Quicktime using the 'export' option. I hope this has not changed its sound. I am currently listening to it in Foobar2000. :righton: :)

[Metoo]

Barry, isn't the dithering supposed to be applied when going to a lower bit depth? If so, what sense does dithering have in this 32-bit context, or are we simply listening to the SRC differences (no dithering yet)?

[Grant]

You do not have to do any dithering when doing SRC, but if you don't, your files will still be at a high bit-depth.

[Cornholio]

Barry, I have just finished downloading your first aiff 32-bit file and find that I cannot play it. I tried playing it with Quicktime and all I get is a kind of shaky hiccupy sound. My other programs do not even recognize the format. :(

Note: I could finally listen to it by converting your aiff file to WAV in Quicktime using the 'export' option. I hope this has not changed its sound. I am currently listening to it in Foobar2000. :righton: :)
Winamp plays all 3 files just fine.

[bdiament]

Barry, isn't the dithering supposed to be applied when going to a lower bit depth? If so, what sense does dithering have in this 32-bit context, or are we simply listening to the SRC differences (no dithering yet)?

Hi Freddy,

Yes, dither would be used when going to a shorter word length (i.e. lower bit depth). It was not used here because all three files on this page are 32 bit float. No dithering has been applied. The page just shows the effects of two different sample rate conversion algorithms. (I always do any processing, including SRC, at 32 float.)

After SRC, I'll apply dither to bring the files down to 16 bit for CD. At some point, I'll put up a similar page with dither examples. On that page, the original will be one of these SRC files (the iZotope converted one).

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

[bdiament]

Winamp plays all 3 files just fine.

Thanks Cornholio,

I was hoping folks on any platform could hear the files.

It will be interesting to hear folks' descriptions of what, if any, differences they hear between the original and the SRC files and between each of the SRC files.

My wife did this last night (bless her patient, music loving heart), without knowing which was which. I just played music for her and switched between the two SRC files. She described the differences she heard (which matched my impressions). Then I asked her to compare each of the two SRC files to a third file, without telling her that was the original, hi-res file. Again she matched my impressions. And interestingly for those who still wonder about whether hi-res offers improvements over 16/44 CD, one of her first comments was how much better defined and more dynamic the bass was. Then she commented about focus, space, treble clarity, etc.

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

[Metoo]

You do not have to do any dithering when doing SRC, but if you don't, your files will still be at a high bit-depth.:confused:

Winamp plays all 3 files just fine.After having converted the files to WAV format I have no problem with Foobar 2000. I don't have Winamp installed because I feel more comfortable with Foobar 2000's feature and interface.

[Metoo]

My experiences last night were as follows:

The following opinions were before knowing which SRCd file was which-
- The first SRCd file (the one done with Peak) sounded quite close to the original at first listen. Yet the original seemed to have more bass presence and more air.
- the second SRCd file (the one done with iZotope) was also quite close to the original. It seemed to have more transient information, which made the first guitar arpeggios to sort of stick out more than in the previous file. It also showed more ambiance information (reverb trails).

Albeit this showing of more detail I had the feeling that the one done with Peak seemed to sound more 'balanced' within itself.

Then I read which was which...

The iZotope, I say again, has more detail. Both SRCd files sound quite close to the original. Perhaps in the case of the iZotope one alongside the more detail there is also some more of punch in the bass. Yet, I am almost sure that some of the 'presence' and 'depth' in both the SRCd files will be lost once they are taken down to 16 bits. It will be interesting to see how much of this will be kept by a good dithering choice.

At the moment I feel that I cannot opine any more on these files because I now know what each of them is and this could affect my opinions.

In any case, given I do not have the ABx module of the Foobar 2000 program installed it has not been easy to compare these files in the program. I also am not sure that the program has not been doing some kind of upsampling to the files.

[Dennis Metz]

I thought Scott Richard had a new band...

[bdiament]

My experiences last night were as follows:

The following opinions were before knowing which SRCd file was which-
- The first SRCd file (the one done with Peak) sounded quite close to the original at first listen. Yet the original seemed to have more bass presence and more air.
- the second SRCd file (the one done with iZotope) was also quite close to the original. It seemed to have more transient information, which made the first guitar arpeggios to sort of stick out more than in the previous file. It also showed more ambiance information (reverb trails).

Albeit this showing of more detail I had the feeling that the one done with Peak seemed to sound more 'balanced' within itself.

Then I read which was which...

The iZotope, I say again, has more detail. Both SRCd files sound quite close to the original. Perhaps in the case of the iZotope one alongside the more detail there is also some more of punch in the bass. Yet, I am almost sure that some of the 'presence' and 'depth' in both the SRCd files will be lost once they are taken down to 16 bits. It will be interesting to see how much of this will be kept by a good dithering choice.

At the moment I feel that I cannot opine any more on these files because I now know what each of them is and this could affect my opinions.

In any case, given I do not have the ABx module of the Foobar 2000 program installed it has not been easy to compare these files in the program. I also am not sure that the program has not been doing some kind of upsampling to the files.

Hi Freddy,

Thanks for your feedback.
I must say that the very first thing I noticed was the added brightening on the file SRCd with Peak. I suspect this might be from the generation of spurious harmonics (i.e. harmonic distortion) in the process. I did not hear this brightening with iZotope.

If you want to hear full Peak SRC + Pow-R3 dithering vs. full zZotope SRC with iZotope dithering, compare these 16/44 files:

http://www.soundkeeperrecordings.com/tracks/sr001-01-1644.wav
and
http://www.soundkeeperrecordings.com/tracks/z03-164401mbmu.aif

The first one is from the Format Comparison page I put up a few weeks ago and is in WAV format. The second was created yesterday and is in AIF format. Sorry for the different formats. I work in AIF. The WAVs were an exception. You should be able to save one to the other without a problem.

You can also hear the original 24/96 version on the Format Comparison page:
http://www.soundkeeperrecordings.com/format.htm

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

[Grant]

:confused: In other words, they do not have to be done together. You can do SRC on a 96k file and render it as 24 or 32-bit 44.1.

[Metoo]

O.K., now I get it. Thanks. :) But, - question for you and Barry - does it make any difference if one does it in two steps or in one go?

[bdiament]

O.K., now I get it. Thanks. :) But, - question for you and Barry - does it make any difference if one does it in two steps or in one go?

Hi Freddy,

I have a program called Sample Manager that lets me "script" processes, such as SRC first, then dither. I don't use this function. My preference is to do each step in the process by itself. I can't say for sure if this produces any benefits beyond my liking it this way.

I never actually process a file either. I make a copy of the file and process the copy. For the next step, I'll copy that processed copy and perform the next process on the new copy. This way, I have every tune for a project saved several times, each one step closer to the finished master. I've found this useful when I want to go back to a particular step, since I don't have to recreate all the prior steps. Also, if something goes wrong during a process (which I have experienced), I don't have to do all the prior steps again. I can just go to the last known good version and proceed from there.

As to actually performing SRC and dithering in a single step, I don't believe this is desirable since all other processes must be completed before dithering (which should always be the very last step). And anyway, as I said, I like to be able to check the results of each step before proceeding to the next.

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

[rnranimal]

Hi Freddy,

Thanks for your feedback.
I must say that the very first thing I noticed was the added brightening on the file SRCd with Peak. I suspect this might be from the generation of spurious harmonics (i.e. harmonic distortion) in the process. I did not hear this brightening with iZotope.

If you want to hear full Peak SRC + Pow-R3 dithering vs. full zZotope SRC with iZotope dithering, compare these 16/44 files:

http://www.soundkeeperrecordings.com/tracks/sr001-01-1644.wav
and
http://www.soundkeeperrecordings.com/tracks/z03-164401mbmu.aif

The first one is from the Format Comparison page I put up a few weeks ago and is in WAV format. The second was created yesterday and is in AIF format. Sorry for the different formats. I work in AIF. The WAVs were an exception. You should be able to save one to the other without a problem.

You can also hear the original 24/96 version on the Format Comparison page:
http://www.soundkeeperrecordings.com/format.htm

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


For the Peak SRC, was the auto-limiter option checked?

[bdiament]

For the Peak SRC, was the auto-limiter option checked?

Hi rnranimal,

No it was not.

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

[Metoo]

Barry, I've been listening to and comparing the original to the two downsampled files with dithering. The Peal SRC one does seem to show some kind of tizz in the sibilants, which the iZotope 64-bit SRC one does not. In fact, this one sounds quite close to the original to me if it weren't because the bass sounds smoother on the original and it shows more air.

[Grant]

O.K., now I get it. Thanks. :) But, - question for you and Barry - does it make any difference if one does it in two steps or in one go?
One should always do the SRC first. You want to preserve the dynamic range up until the very last moment. Besides, you must dither at the very last moment too. If you dither before you SRC, your SRC will not sound good. You will increase the noise by processing a dithered file, so there' sno need.

As you know, or should know, if you ever do need to go back and process an already dithered file, it's best to not use any dither to avoid double dithering. Not only will you get more noise by a double dose, but you will smear the sound. If you don't use dither upon processing a second time, your sound will be a bit thinner, but the dynamics will pretty much be preserved, and transients will be sharp. And, this way, you may be able to get away with processing the 16-bit file. But, you don't even deal with 16-bit, so...:)

Barry, I don't save every step, but I do save the orignal hi-rez files. I just note my settings.

[bdiament]

...As you know, or should know, if you ever do need to go back and process an already dithered file, it's best to not use any dither to avoid double dithering. Not only will you get more noise by a double dose, but you will smear the sound. If you don't use dither upon processing a second time, your sound will be a bit thinner, but the dynamics will pretty much be preserved, and transients will be sharp. And, this way, you may be able to get away with processing the 16-bit file. But, you don't even deal with 16-bit, so...:)

Barry, I don't save every step, but I do save the orignal hi-rez files. I just note my settings.

Hi Grant,

What you've said is one reason I'm very glad I have a copy of each file saved at each step along the way. I can go back to the original, an EQd version, a gain adjusted one, one with a fade applied or with special edits, one will all these processes, a 32 bit or 24 bit non-dithered 96k, a 32 bit SRCd 44k and more. I never have to perform any process a second time and can instantly return to any step along the way.

If I didn't keep all these copies and I had to do more work after dithering a file, I'd go back to the original and start over again before I performed any process at all on a dithered file.

Dithering, as you know, must be the very last step. Any process after dithering defeats the purpose as you are back to truncating new detail (any process will lengthen the word and those extra bits will be lost without dither). You have the disadvantage of the added noise -from the previous dither- without the benefits.

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

[Grant]

Dithering, as you know, must be the very last step. Any process after dithering defeats the purpose as you are back to truncating new detail (any process will lengthen the word and those extra bits will be lost without dither). You have the disadvantage of the added noise -from the previous dither- without the benefits.

That's right! But, I have, on occasion, had to work on something that had been dithered, with no backup available. So, in those cases, it's good to know how you can proceed and do the least amount of damage to the sound.