I am looking for the best way to resample some vinyl rips for cd burning. I've tried the dbpoweramp nobrainer way, but I was thinking that Audition or Soundforge might produce better results. Could you be as specific as possible in terms of settings and such? Any links to any guides would also be appreciated.

Thanks.

Try the PPHS resampler plug-in in Foobar2000 (one of the DSP's). You can set it up to be active when you convert.

Voxengo R8Brain is possibly the best quality free option. http://src.infinitewave.ca/

"quality free"

1. pphs in foobar or
2. ssrc


later

That looks like a very helpful link, but I'm not sure how to interpret the graphs...I've always used Adobe Audition w/default settings to resample 24/96 to 16/44.

If cost of software is no object, what is the best software/method/settings to use to resample?

Cost no object? Voxengo R8Brain Pro. It's not actually that expensive but it is the best. See graphs and recommendation by Bob Katz

This paper may help make the infinitewave site graphs more meaningful
http://www.bias-inc.com/products/peakPro5/...gWhitePaper.pdf

Thanks, that's very useful

I'll try some tests on my own when I get some free time to see how my results compare.

I love the marketing speak at the top of that document...

Not to mention Adobe Audition, which in those test appears to provide equally good quality, and has been doing so (in the previous incarnation of Cool Edit / Pro) for over a decade!

Interesting that http://src.infinitewave.ca/ shows artefacts with Adobe Audition that I've never noticed before. The greatest spectral range I've ever used is 140dB, and that allows me to see the bottom of the noise shaped dither on 16-bit recordings. Switching to 200dB reveals the same artefacts from Cool Edit Pro that http://src.infinitewave.ca/ shows from Adobe Audition.

e.g. see this image of the result of resampling an 8 second tone sweep from 96kHZ to 44.1kHz (both at 32-bits):
Click to view attachment

To put this into perspective, if target is actually 44.1kHz at 16bits, I don't think the "artefacts" will be audible through the noise floor...
Click to view attachment

Cheers,
David.

Ok, I've tested internal Audition's converter and popular foobar plugins (PPHS from 0.9.4.3 Standard DSP array, SSRC v0.57 and Secret Rabbit Code v1.03). All were using their top quality settings (PPHS and SSRC Ultra mode, Rabbit Best Sinc Interpolator and Audition with Pre/Post Filter and Quality set to 999). Sweep is generated in Audition as 96kHz/32-bit. All converted files were also left in 32-bit float.

Resolution was set to 512 bands, windowing function used is Blackmann-Harris with 75% window width. Range is set to 180dB.

Audition:


PPHS:


SSRC:


Secret Rabbit Code:


As David said, when the files are converted to 16bit with 1bit dither, artefacts are covered with the noisefloor in all except the Secret Rabbit. SSRC also have some problems at highest frequencies (which is also visible in the time domain—the amplitude goes up a little). Audition clearly wins.

Just download R8Brain Free. http://www.voxengo.com/product/r8brain/

It is the proven best free method of resampling. It also comes with a "Reusable royalty-free converter DLL with explained API"
For my money, Voxengo make the best damn sounding software to date. This guy knows what he is doing.

eevan, could you show me how to do these test? There are a few things I would like to test myself.

Thanks!

For Audition/CoolEdit, the inside word for years has been that setting the quality slider above 250 accomplishes nothing except to lengthen the processing time significantly. The result of the increased calculation precision with higher settings is far below audibility.

OK. You need a test file. I have generated this one using Audition's Generate Tones tool. You set up the final freq of the sweep in the 'Base frequency' and use the same value for 'Modulate By'. Then you set the 'Duration' and based on that set the 'Modulation Frequency' to 1/Duration (one full cycle up and down around the Base Frequency). Now you just cut off the unwanted part of the resulting wave.
Now you use this file as an input to whatever you want to test.

It's best to do all this on 32bit float file. If your test object can't hande this type, convert the file to supported integer format.

Hope this helps!


You are right Andy, I'm aware of that. Just wanted to push every setting to the limits

I don't know what you intend by "Now you just cut off the unwanted part of the resulting wave." but you need the sweep tone to go well above the Nyquist limit of your target sample rate in order to demonstrate the most obvious artifact, the alias image. Of course there will not be one with the proper settings in Audition, but lesser programs will have one.

I'll clarify this.
In the tests above I needed a sweep from 0–48000Hz (all frequencies that can exist in a file sampled at 96khz). You set the base frequency to 48kHz (let say the carrier for frequency modulation) and modulate by 48kHz (in fact, 100% modulation factor). The modulation is done with sine wave. If you setup the Generator as I explained, during the first half of the resulting wave you'll get the silence (because the frequency of the carrier is supposed to go to 96kHz and back to 48kHz, and that is impossible at 96kHz sample rate). In the other half of the resulting wave first you get a sweep from 48kHz down to 0 (1Hz actually), followed by a sweep back to 48kHz. For the sake of clearer spectrograms, I wanted only one sweep. So I left only the final one (from 0–48kHz) and cut off everything else. It would be much easier if one could set the start phase of the modulating wave, or have in Audition a dedicated built-in sweep generator.

Here's the snapshot:


Cheers

Perhaps most people are more adept in the language and can thus follow your instructions but I, alas, am missing something. I don’t understand how to follow the instructions in your post.

There does seem to be a possibly more straightforward approach, however. I always generate a sweep frequency by setting the lower frequency under the Initial Settings tab and the upper frequency under the Final Settings tab. Of course, one could go the other way around.

One can enter whatever suits under Modulate By, Modulation Frequency, and in the Phasing boxes. I don’t see any general need for those values to be anything but zero, but perhaps I’m missing a trick.

There is no need to edit the generated tone file.

Maybe eevan has missed the initial/final settings and is attempting to simulate them using the modulate option?

Or maybe the particular sweep rate isn't available with the initial/final settings options, and can only be achieved using modulate?

(I think the latter is true. The CEP "log sweep" didn't look like the plots on the website. I just used linear sweep).

Cheers,
David.

Using Initial and Final settings is ok for linear or logarithmic sweep. This method allows you to make a sweep according to sine function. Although it's not so commonly used, I wanted visually to reproduce the graphs from http://src.infinitewave.ca/ which seems to follow the sine function.

Of course, I know it's much easier to get lin or log sweeps.

Thanks for the voxengo suggestion, quite helpful.