Converting various sample rates and bit resolutions without subjective quality loss

[Zownd]

your order page goes blank.

tried "chrome" and "safari"

[sanddigger1]

Same with Firefox.

removing quantization errors coming from previous processing stages, what is done by converting to the same frequency and the same bit depth.

What does that mean ?

[Twerk]

The name iSRC is a bit confusing as we already have such a thing, and it's completely unrelated to sample rates. You might consider something a bit more unique.

Thanks for sharing, I'll check it out.

[Helios67]

Website has been finished yesterday evening and it seems was not working properly because of inserting rate calculator. Improved and should work fine now. Sorry for troubling.
iSRC stands for "Integer Sample Rate Converter" and this is what is that. Might be sort of misleading because of "International Standard Recording Code", that is why we used to name it iSRC, not ISRC. Probably not enough difference. It is a brand fresh thing, so the highest time to change something. Thanks for a lightning.
We will think about it and will be updating somehow. Anyway the program is pretty useful, might work better for sound than tons of plugins, so feel free to check it out.

[Helios67]

Quote:
removing quantization errors coming from previous processing stages, what is done by converting to the same frequency and the same bit depth.

What does that mean ?

You can hear a difference yourself when you just convert your files for the same sampling frequency and resolution. I mean it works good for changing formats or just processing the formats which don't need to be changed. Also you may do some tricks like increasing frequency with slight reduction of resolution. And go back to the format you need. It may bring back more music to your files. This is a tool and you are the master.

[Helios67]

Any opinions to share?

[Zownd]

Originally Posted by Helios67

Any opinions to share?

Well, I am not sure why I would pay such an amount compared to a tried an tested package like Barbabatch. I would have expected something in the $29 range. Not a tenfold of that. Sorry for my honesty here

Also I think it's better to timebomb a package instead of 1 minute snippets.

Is there any other software you are known for?

Cheers,
Mark

[Greg Reierson]

Originally Posted by Twerk

The name iSRC is a bit confusing as we already have such a thing, and it's completely unrelated to sample rates. You might consider something a bit more unique.

That's what I was thinking. The last thing the industry needs is more confusion about ISRC and/or SRC.

GR

[Helios67]

Originally Posted by Zownd

Well, I am not sure why I would pay such an amount compared to a tried an tested package like Barbabatch. I would have expected something in the $29 range. Not a tenfold of that. Sorry for my honesty here

Also I think it's better to timebomb a package instead of 1 minute snippets.

Is there any other software you are known for?

Cheers,
Mark

This is sort of basic misunderstanding. There are dozens of programs (each studio system may do it easily) enabling you to make any digital conversion. The point is HOW IT IS DONE. I mean when higher formats/resolutions are converted to the lower ones, the sound is getting collapsed in more or less significant way. All currently available programs have a problem with that. This was the reason we reasearched the issue during the years and tried to design the tool making this in a RIGHT way. We think we did the work with great results. That is why we decided to release it commercially. Nykiel iRSC works great even when you don't want to change the format as it covers previous quantization errors. The result is your sound is getting more open, musical, fluent and set up unless you have really excellent source material. The program cannot be comparable to the industry standard as it is far more advanced. Nykiel iRSC is for those who want to preserve highest possible quality throughout the whole digital processing. I hoped some members of mastering forum should be interested. And some probably will do sooner or later. We don't have any other programs commercially available as yet.

[dcollins]

Originally Posted by Helios67

This is sort of basic misunderstanding. There are dozens of programs (each studio system may do it easily) enabling you to make any digital conversion. The point is HOW IT IS DONE. I mean when higher formats/resolutions are converted to the lower ones, the sound is getting collapsed in more or less significant way. All currently available programs have a problem with that.

How is yours different? Do you have any objective data to show?

You should reproduce the tests from the infinitewave site, or submit your system to them to add to the other tests.

Suffice it to say, there are implementations that are essentially 'perfect' out there already. Maybe yours is more perfect?

The program cannot be comparable to the industry standard as it is far more advanced.

But maybe it can....................


DC

[Zownd]

Originally Posted by Helios67

This is sort of basic misunderstanding. There are dozens of programs (each studio system may do it easily) enabling you to make any digital conversion. The point is HOW IT IS DONE. I mean when higher formats/resolutions are converted to the lower ones, the sound is getting collapsed in more or less significant way. All currently available programs have a problem with that. This was the reason we reasearched the issue during the years and tried to design the tool making this in a RIGHT way. We think we did the work with great results. That is why we decided to release it commercially. Nykiel iRSC works great even when you don't want to change the format as it covers previous quantization errors. The result is your sound is getting more open, musical, fluent and set up unless you have really excellent source material. The program cannot be comparable to the industry standard as it is far more advanced. Nykiel iRSC is for those who want to preserve highest possible quality throughout the whole digital processing. I hoped some members of mastering forum should be interested. And some probably will do sooner or later. We don't have any other programs commercially available as yet.

Hey! don't shoot the messenger !!
You asked, so I replied.

Barbabatch is an award winning application that has been around for quite some time.
They improved on every release going back more than ten years.

Off course we are interested in an application like this, but don't forget that our audioworkstations are getting better with each release as well.
So for a lot of conversions we are talking minor differences here.
I guess you should start by getting it to the bigger magazines. I can and will test and review it for you for the biggest magazine here in the Netherlands and probably also "Sound on Sound" so drop me a line if you are interested.

Sadly our music will end up on i-tunes as a bad mp3 anyway

Cheers,
Mark

[Helios67]

Originally Posted by dcollins

How is yours different? Do you have any objective data to show?

You should reproduce the tests from the infinitewave site, or submit your system to them to add to the other tests.

DC

Our Nykiel iSRC is based on different way of interpolation and decimation calculation. And different approach to handle requantization errors optimized due to the short-term temporary errors (modulation errors). That is very "objective" as you can hear that. Others "objective" data are not that much important as in fact they tell you nothing about what REALLY matters. PLEASE NOTE THAT ALL VINTAGE STUFF LIKE TELEFUNKEN ELA M 250/251, AKG C12, NEUMANN U47, MAIHAK V41, FAIRCHILD 670 TO NAME A FEW ARE "OBJECTIVELY" WORSE THAN ANY STUDIO MODERN EFFORTS. You can measure vintage gear with good results, but it didn't tell you why it sounds SO GOOD. So let's believe our ears and musicality. If you hear it is working well, it just means that. If not, just pass. You don't need John Lennon (it would be nice) to tell you about. Simple as that.

[Helios67]

Originally Posted by Zownd

Off course we are interested in an application like this, but don't forget that our audioworkstations are getting better with each release as well.

I dare to doubt unfortunately. The updates are focused about logistic and flexibility, not the sound itself as basic approach is not changed. Anyway I am analog head and do not want to discuss about that. [/QUOTE]

Originally Posted by Zownd

So for a lot of conversions we are talking minor differences here.
I guess you should start by getting it to the bigger magazines. I can and will test and review it for you for the biggest magazine here in the Netherlands and probably also "Sound on Sound" so drop me a line if you are interested.

We think differences are not minor ones. This is not only my opinion. And that's the point. Would recommend to make some testing.
Obviously we are thinking about contact with MIX and other magazines. Thanks for your offering. Surely will contact you.

Originally Posted by Zownd

Sadly our music will end up on i-tunes as a bad mp3 anyway

Sure, this is very sad. However there are a couple of milions folks having better and better hi-fi and they are starving for really good sound above industry standard. And they are still buying records. Let's do not forget about them.

[Zownd]

Originally Posted by Helios67

Sure, this is very sad. However there are a couple of milions folks having better and better hi-fi and they are starving for really good sound above industry standard. And they are still buying records. Let's do not forget about them.

definitely not!

Please PM me or mail me about the mags.

Cheers,
Mark

[Alexey Lukin]

Originally Posted by Helios67

Our Nykiel iSRC is based on different way of interpolation and decimation calculation. And different approach to handle requantization errors optimized due to the short-term temporary errors (modulation errors).

You guys go great lengths to prove that 32-bit floating point FIR filtering is inappropriate for SRC, but you fail to recognize a simple fact that your competitors don't typically use 32-bit FP accumulators. Even when they write this:

	for( uint i=0; i<nSignalPoints; ++i ) {
		float fSum= 0;
		for( uint j=0; j<nTaps; ++j )
			fSum+= x[i+j] * y[j];
		z[i]= fSum;
	},

a typical optimizing compiler will do the summation in 64-bit double precision, either on FPU or on SSE2.

[Bob Olhsson]

Originally Posted by dcollins

...

But maybe it can....................

The paper seems to reach the very same conclusions Andy Moorer did during the late '90s.

[PBM]

Originally Posted by Helios67

Our Nykiel iSRC is based on different way of interpolation and decimation calculation. And different approach to handle requantization errors optimized due to the short-term temporary errors (modulation errors). That is very "objective" as you can hear that. Others "objective" data are not that much important as in fact they tell you nothing about what REALLY matters. PLEASE NOTE THAT ALL VINTAGE STUFF LIKE TELEFUNKEN ELA M 250/251, AKG C12, NEUMANN U47, MAIHAK V41, FAIRCHILD 670 TO NAME A FEW ARE "OBJECTIVELY" WORSE THAN ANY STUDIO MODERN EFFORTS. You can measure vintage gear with good results, but it didn't tell you why it sounds SO GOOD. So let's believe our ears and musicality. If you hear it is working well, it just means that. If not, just pass.

Suddenly - the scintilla of a hint of a piscine pong ...

[Helios67]

Originally Posted by Alexey Lukin

You guys go great lengths to prove that 32-bit floating point FIR filtering is inappropriate for SRC, but you fail to recognize a simple fact that your competitors don't typically use 32-bit FP accumulators. Even when they write this:

	for( uint i=0; i<nSignalPoints; ++i ) {
		float fSum= 0;
		for( uint j=0; j<nTaps; ++j )
			fSum+= x[i+j] * y[j];
		z[i]= fSum;
	},

a typical optimizing compiler will do the summation in 64-bit double precision, either on FPU or on SSE2.

Even if you believe they do this in a right way, Nykiel iSRC did this with different approach and much better results for the ears. I recommend to make some tests. if you don't get a better sound, no point to use it.

[bandpass]

Originally Posted by Alexey Lukin

You guys go great lengths to prove that 32-bit floating point FIR filtering is inappropriate for SRC

Insufficient internal precision would result in noise being evident in the output; but in fact, this new src seems to have a worse noise-floor, even than float32 srcs.

Here compared against the float32 version of ssrc (with TPDF amplitude 1):

[Helios67]

[QUOTE=bandpass;7664174] this new src seems to have a worse noise-floor, even than float32 srcs.
/QUOTE]

Sure, this is the point. Nykiel iSRC is carefully balanced to get the best musical effect in digital domain. If you are pressing to get "objective" parameters as high as possible the sound is getting rather worse than better. We are not able to participate in any "parameters" race as it goes to the place we left and we cannot see any reason to go back. So please let's talk about your real aural experiences, if you need such a tool or you are just happy what you already have.

[Jon Hodgson]

So, it has no subjective effect on what you hear... unless there's a "quantization error" in the signal you feed into it... in which case it DOES have a subjective effect?

How does it tell the diference between signal components created by quantization and those not, and then remove them, without any effect on anything else?

[Laarsø]

Originally Posted by Alexey Lukin

You guys go great lengths to prove that 32-bit floating point FIR filtering is inappropriate for SRC, but you fail to recognize a simple fact...
a typical optimizing compiler will do the summation in 64-bit double precision, either on FPU or on SSE2.

Please clarify the terms. Seeing that 64 bit = 32 bit double-precision. 64 bit double-precision = 128-bit. How many SHARCs you got in that tank?


Dr. Moorer was not against floating-point maths. He was against only using 32 bits. He wrote in his white paper on why 48 bits beats 32 that he'd recommend 64 bits (floating point) but that, at the time of writink, only MathLab was usink 64 bits - too "expensive" for practical audio applications. Also, the fixed-point maths was lettink him use idealized state variable filters - I believe, only for EDL strip equalization, rather than Melchoir's(?) SFC - than those which come automagically for floatink point, but which are not as precise.



Just sayink...



Cheersø,
Laarsø

[Laarsø]

Originally Posted by Greg Reierson

That's what I was thinking. The last thing the industry needs is more confusion about ISRC and/or SRC.

GR

Yes, the name is a "bit" confusink. Seeing that the application is not only for converting sampling rates, but also for converting sample format, and error-deletion?, I suggest the followink rename:


iC. This stands for "integerConversion." The things this application does are all basically a form of fixed-point math sample conversion. Sampling rate iss just one of things that your application changes. Also, iC doesn't read like the "ISRC," or "iirc," etc....

Another rename idea might be, {Z}Con, which would stand for Zahlen-Converter.


Lots of possibilities which won't confuse we blotto premasterink clerks.


Cheersø,
Laarsø

[Alexey Lukin]

Originally Posted by Laarsø

Please clarify the terms. Seeing that 64 bit = 32 bit double-precision. 64 bit double-precision = 128-bit. How many SHARCs you got in that tank?

64-bit FP format is called "double precision".
32-bit FP format is called "single precision".
SHARCs are not considered in my post, although it's quite common for DSPs to have an extended-precision (integer) accumulator for summing long dot products.

[Alexey Lukin]

Originally Posted by bandpass

but in fact, this new src seems to have a worse noise-floor, even than float32 srcs.
Here compared against the float32 version of ssrc (with TPDF amplitude 1):

It's very likely that SSRC uses a 64-bit accumulator, i.e. is not purely a 32-bit SRC. See my code example above.

[Helios67]

Here is a short note from Peter Nykiel, who designed iSRC. It gives more technical details for those who are interested:
"Our sample rate converter employs an unique approach for minimizing conversion errors. Instead of time-invariant models describing such phenomena as the polyfilter mismatch in the process of interpolation or the emergence of numerical and quantization errors which are usually modeled using additive stationary noise sources, more realistic, nonlinear and time-dependent models have been used. The proposed models assume the error is directly correlated with the signal transients and it temporarily changes the properties of the process, thus making it time-variant. So, in our approach such changes in converter's transfer function have been reduced as much as possible in the audio frequency band. The subjective tests confirm high quality and fidelity of the conversion compared to any other solutions available on the market. So, in our method short-time, time-variant distortions are minimized as much as possible. Long-time, objective parametrs such THD+N do not even see time-variant distortions, bacause they rely on analysis of long-term periodic test signals."

And to my analog ears it just sounds right.

[Helios67]

Originally Posted by Laarsø

Please clarify the terms. Seeing that 64 bit = 32 bit double-precision. 64 bit double-precision = 128-bit. How many SHARCs you got in that tank?


Dr. Moorer was not against floating-point maths. He was against only using 32 bits. He wrote in his white paper on why 48 bits beats 32 that he'd recommend 64 bits (floating point) but that, at the time of writink, only MathLab was usink 64 bits - too "expensive" for practical audio applications. Also, the fixed-point maths was lettink him use idealized state variable filters - I believe, only for EDL strip equalization, rather than Melchoir's(?) SFC - than those which come automagically for floatink point, but which are not as precise.

http://nykiel-audio.pl/wp-content/up...ating-fmt2.pdf

[Alexey Lukin]

Originally Posted by Helios67

Instead of time-invariant models describing such phenomena as the polyfilter mismatch in the process of interpolation or the emergence of numerical and quantization errors which are usually modeled using additive stationary noise sources, more realistic, nonlinear and time-dependent models have been used. ... So, in our method short-time, time-variant distortions are minimized as much as possible.

The distortions that you ascribe to "other solutions" are virtually nonexistent. If peak distortions are below -90 dB, you do not need models to evaluate them — they just don't matter.


I generally agree with this sentence:

Originally Posted by Helios67

The subjective tests confirm high quality and fidelity of the conversion compared to any other solutions available on the market.

but it does not state that iSRC is superior. It seems to be just as high in quality as other competing products.

[Helios67]

Originally Posted by Alexey Lukin

The distortions that you ascribe to "other solutions" are virtually nonexistent. If peak distortions are below -90 dB, you do not need models to evaluate them — they just don't matter. ...
...it does not state that iSRC is superior. It seems to be just as high in quality as other competing products

They matter more than you think even if you don't understand or disregard them. It is a innovatory research not following industry standard we were not satisfied with. This was the reason to work out a new approach. If you are satisified with the program you already have, you probably just don't need anything else. Anyway you can always make a test and make some experiments to hear how it works. Simple as that.