SRC Comparison Website

[IxtabStudios]

I think this is the right forum...

I came across this interesting site on SRC Comparison tests:

SRC Comparisons

My question is how do these pretty pictures sound? From looking at the graphs, iZotope RX seems to be the best (better than Weiss looking at the picture), but does that mean it sounds good?

From the graphs I'd say Cubase or DP shouldn't even have downsampling as an option!

Just looking for the best sound possible...

[peeder]

I use the izotope SRC, and have no complaints...

[Eide]

so according to this webpage - which is the best one? i cant completely understand these diagrams

[johnlink]

Be sure to press the HELP button for explanation of the graphs.

John Link

[MAzevedo]

Well, for the Passband, Transition, and Phase responses, the white line represents 'perfect' response, and the green line is the response of the algorithm being tested.

For the sweep test, a perfect result would be a white line going from zero hertz to 22k5Hz, and then coming to a dead stop and the rest of the graph being pure black.

For the 1kHz test, the ideal result would be a single pure spike at 1kHz and nothing else.

One thing I wonder though, the 'perfect' response in the Phase test is described as a linear-phase filter. Would pre-ringing show up on that graph, or does it assume that a linear-phase filter (with the potential for pre-ringing) always be preferred to a minimum-phase filter?

[theblotted]

iirc, someone once said that the graphs are made by iZotope ppl... so not sure how biased the graphs are..

otoh, i do use iZotope's algo for SRC, it's been working fine.

[jayfrigo]

Originally Posted by IxtabStudios

I think this is the right forum...

I came across this interesting site on SRC Comparison tests:

SRC Comparisons

My question is how do these pretty pictures sound? From looking at the graphs, iZotope RX seems to be the best (better than Weiss looking at the picture), but does that mean it sounds good?

From the graphs I'd say Cubase or DP shouldn't even have downsampling as an option!

Just looking for the best sound possible...

The pictures don't tell you everything; for instance they don't indicate pre-echo/ringing in any graphs. And even what they do tell must be interpreted. First, look at the more than just the sweep. The Weiss Saracon is by far the cleanest on the 1 kHz tone graph, though the iZotope and some others also perform very well.

As for the sweep, Weiss actually optimizes the filter to intentionally leave a tiny bit of aliasing at very high frequencies where it should be practically impossible to hear, as a tradeoff to minimize echo/ringing which you are more likely to hear. As a result, it sounds better even though the sweep shows a tiny bit up top. This is where understanding SRC and interpretation of results comes in. What you see is actually a good thing when it comes to how it sounds.

iZotope now allows you control of filters so that you can achieve a similar balance between aliasing and pre-echo. I love that they opened up the control of both the steepness of the filter, and balance between FIR and IIR (linear and minimum phase). The SRC in RX Advanced also sounds fabulous. While I feel that the Weiss is still the one to beat, the iZotope is nipping at its heels. For practical purposes, You probably could not tell them apart in a blind test under most circumstances. Both are excellent.

[Alexey Lukin]

Originally Posted by MAzevedo

One thing I wonder though, the 'perfect' response in the Phase test is described as a linear-phase filter. Would pre-ringing show up on that graph, or does it assume that a linear-phase filter (with the potential for pre-ringing) always be preferred to a minimum-phase filter?

No, it is not implied that a linear-phase filter is perfect. The straight line is just for reference. Similarly, it's not implied that the steepest filter is the best: higher steepness always means more time-domain ringing. So, there's no single perfect filter, and it can be subjective.

[Eide]

so according to the website Izotope is the winner?

[MAzevedo]

Originally Posted by Alexey Lukin

No, it is not implied that a linear-phase filter is perfect. The straight line is just for reference. Similarly, it's not implied that the steepest filter is the best: higher steepness always means more time-domain ringing. So, there's no single perfect filter, and it can be subjective.

In the Help pages, I found that there is a graph of impulse response, which shows the pre-ringing of the linear phase filters. It's too bad there isn't a impulse response for each filter among the other tests they have, I think that's the only thing that is really missing.

[Alexey Lukin]

That's true, and maybe such graphs will be added in the future. However there's a very credible way to estimate how the impulse response looks - just by looking at the Transition graph. The steepness of the filter cutoff is directly linked to the number of oscillations in the impulse response. And whether it's a post- or pre-/post-ringing, can be estimated from the shape of phase response. Currently there were only 2 types of filters: linear-phase (equal pre- and post-ringing) and minimum-phase (only post-ringing), except iZotope "intermediate-phase" filters, which can balance pre- and post-ringing to some extent.

[jayfrigo]

Originally Posted by Eide

so according to the website Izotope is the winner?

No. Go back and re-read this thread. It seems you must have missed a thing or two.

The point of that site is not to proclaim a winner, and even if you wish to, while iZotope is in the running, it couldn't be called the clear winner. A few others meet or exceed it in certain properties. The iZotope is definitely one of the good ones, but not the only one.

[Coyoteous]

Is the iZotope SRC in Wave Editor 1.3 the same as that in iZotope RX Advanced?

WE 1.3 has advanced options with controls for Filter Steepness (0-200), Cutoff Scaling (.5-2) and Preringing (0-1) as well as an overall Quality Setting (low-high). It also has Presets for Gentle, Linear Phase / Intermediate Phase / Minimum Phase / Steep, Linear Phase / Steep, No Aliasing / Ultra-steep, Linear Phase.

(though I don't understand them yet - thanks to Jay for the expanations above)

[Alexey Lukin]

Originally Posted by Coyoteous

Is the iZotope SRC in Wave Editor 1.3 the same as that in iZotope RX Advanced?

Yes, it's the same algorithm. However RX Advanced currently has a wider range of Steepness (up to 2000).

[Alexey Lukin]

Originally Posted by MAzevedo

It's too bad there isn't a impulse response for each filter among the other tests they have, I think that's the only thing that is really missing.

Good news! The impulse response graphs have been added to SRC Comparisons. Some of the graphs are truncated in time, because certain impulses can obviously be quite long or even infinite.

[MAzevedo]

Originally Posted by Alexey Lukin

Good news! The impulse response graphs have been added to SRC Comparisons. Some of the graphs are truncated in time, because certain impulses can obviously be quite long or even infinite.

Dang. Man, I guess I should ask for more stuff on this forum, that's the best service I've ever gotten on anything. ^_^

Umm.... how about..... I think it would be really great is the SRC Comparison site got me a 3G 16g iPhone next week.

Or a pony.

Whichever is easier guys, no rush or pressure or nuttin'.

[MAzevedo]

And now that I've gone back and looked through the graphs...

I definitely feel better about some of the algorithms that looked 'ugly' before the impulse response tests. For instance, the Secret Rabbit (Linear) conversion has hands down the best impulse response there (and is phase-linear), but the trade off is a fugly sweep and distortion result. Now that the last piece of the filtering puzzle is in place, you can finally see the tradeoffs between distortion in the time, phase, and frequency domains all at once. Woot.

[jayfrigo]

Thanks Alexey. Good stuff.

[Hane]

The SRC graphs proves one thing. In sample rate conversion, you don't get something for nothing. The impulse graphs of some of the software looks excellent, but it's at the expense of other parameters.

[jayfrigo]

Originally Posted by Hane

The SRC graphs proves one thing. In sample rate conversion, you don't get something for nothing. The impulse graphs of some of the software looks excellent, but it's at the expense of other parameters.

It's true. Each parameter individually can be made to look nearly perfect at the expense of the others. The trick is getting just the right balance among them all that translates into something that sounds good in addition to measuring well. That's why it's hard to look at one graph in isolation and try to pick a "winner."

Besides, how can you pick a winner in sound by simply looking at pictures? Don't you think listening should be part of the evaluation process?

Still, wonderful resource the guys have provided. That goes without saying.

[peeder]

Originally Posted by jayfrigo

Besides, how can you pick a winner in sound by simply looking at pictures? Don't you think listening should be part of the evaluation process?

Part of, yes, all of, no. A lot of people think that listening should be all we do and nothing else. That is a mistake, because it will be relatively easy to select material that a given processor's flaws harmonize with well, leaving most material to suffer.

It's just like those demos of MIDI pianos. They get someone to play something that happens to sound awesome on that phony tinny POS piano, and when you get the thing in your studio, you wonder why you ever bought it.

[jayfrigo]

Originally Posted by peeder

Part of, yes, all of, no. A lot of people think that listening should be all we do and nothing else. That is a mistake, because it will be relatively easy to select material that a given processor's flaws harmonize with well, leaving most material to suffer.

I hear ya. It's just that these threads sometimes get pretty heavy on the visual and I thought it might be useful to remind ourselves why we care in the first place!

[Coyoteous]

Originally Posted by Alexey Lukin

Yes, it's the same algorithm. However RX Advanced currently has a wider range of Steepness (up to 2000).

Thanks, are you one of the iZotope principals?

I'm starting to catch on and have found these resources helpful:

http://www.izotope.com/tech/src

http://www.izotope.com/support/help/rx/index.html

http://www.bias-inc.com/products/peakPro5/resampling

http://www.bias-inc.com/products/pea...WhitePaper.pdf

I think Wave Editor and Peak are probably the best SRC's I own. The WE manual has no info about the advanced SRC features (that I can find), though I think I can fill in the blanks between this thread and the first 2 links (above).

Lately, I've been using Peak, which seems to sound good. Set to highest quality, it takes about 10 times realtime on my 1.5GHz Power Book - but it also batches, so overnight jobs are not too much of a problem.

So, if Saracon is the one to beat, what settings in WE would get me closest to that?

[Alexey Lukin]

Originally Posted by Coyoteous

are you one of the iZotope principals?

I'm the algorithm developer at iZotope.

Originally Posted by Coyoteous

The WE manual has no info about the advanced SRC features (that I can find)

There's a screenshot of advanced features at page 94 of their manual: http://audiofile-engineering.com/sup...1.x_Manual.pdf
For explanation of the controls refer to page 56 of the restoration guide: http://izotope.fileburst.com/guides/..._Guide_v_1.pdf

Originally Posted by Coyoteous

So, if Saracon is the one to beat, what settings in WE would get me closest to that?

Steepness = 38, cutoff scaling = 1, pre-ringing = 1.
In the demo of RX Advanced you can change the parameters and watch the plot of frequency response of the SRC filter - to match it to any other SRC you like.

[peeder]

Originally Posted by Alexey Lukin

I'm the algorithm developer at iZotope.

And I thank you for your leadership!

[Coyoteous]

Originally Posted by Alexey Lukin

I'm the algorithm developer at iZotope.

There's a screenshot of advanced features at page 94 of their manual: http://audiofile-engineering.com/sup...1.x_Manual.pdf
For explanation of the controls refer to page 56 of the restoration guide: http://izotope.fileburst.com/guides/..._Guide_v_1.pdf

Steepness = 38, cutoff scaling = 1, pre-ringing = 1.
In the demo of RX Advanced you can change the parameters and watch the plot of frequency response of the SRC filter - to match it to any other SRC you like.

- thanks, Alexey! I think iZotope is my new most favorite company, though I don't have any products directly - just the licensed technology in Wave Editor.

I'm a poor struggling artist and bottom-feeding audio worker these days, so I've got to get by with what I've got. I'm downloading the RX Advanced demo as I type this.

[Jerry Tubb]

Originally Posted by Coyoteous

I'm a poor struggling artist and bottom-feeding audio worker these days, so I've got to get by with what I've got.

Make that ~Ace Audio Detective~ Coyoteous Barry!

Don't know if you realize it, but you're famous around our place for discovering the WaveBurner bit-bug.

I've been very pleased with the "default" iZotope SRC results in Sample Manager. With client budgets being a little tighter this summer, thanks to the greedy oil price speculators on Wall Street, not much extra time in sessions for experimenting with additional SRC parameters.

Major Props to Alexey Lukin for his excellent work.

Best - JT

[Alexey Lukin]

Thank you, we are all working hard at iZotope to create innovative products.

[Coyoteous]

Originally Posted by Jerry Tubb

Make that ~Ace Audio Detective~ Coyoteous Barry!

Don't know if you realize it, but you're famous around our place for discovering the WaveBurner bit-bug.

Darn it, Jerry - you would have to go and spoil my self pity party... and it's good to be famous somewhere, I sure ain't around here!

I've been very pleased with the "default" iZotope SRC results in Sample Manager. With client budgets being a little tighter this summer, thanks to the greedy oil price speculators on Wall Street, not much extra time in sessions for experimenting with additional SRC parameters.

I may get SM at some point. I'd guess that's the same conversion as WE 1.2, which sounds quite good to me - maybe the same as the default setting in WE 1.3?

Major Props to Alexey Lukin for his excellent work.

Best - JT

- indeed and best to you, too!

[Matthew Foust]

Originally Posted by Coyoteous

I may get SM at some point. I'd guess that's the same conversion as WE 1.2, which sounds quite good to me - maybe the same as the default setting in WE 1.3?

FYI, Sample Manager 3.2 will sport advanced options for MBIT+ Dither and the iZotope 64-bit SRC.