I'm recording from vinyl, and I want to store the rip as both 24bit/96kHz FLAC and also as 44.1kHz MP3 (LAME -V0).

Once I have the 96kHz FLACs, am I better off ripping the vinyl again using 44.1kHz sample rate, or would using foobar2000's FLAC->MP3 converter with PPHS re-sampling DSP suffice?

Resampling in foobar2000 is fine. PPHS is considered transparent (and has modest processor cost, which is great for live playback or speedy encoding). SSRC resampler in Ultra mode is perhaps theoretically superior, but slower, and probably has negligible benefit when encoding to lossy.

Including the resampler in the Converter's own DSP chain and passing 24-bit data to LAME is an excellent approach.

If you were to do any processing (declicking etc.) this also saves doing it twice!

One of the best resamplers is Voxengo R8Brain, it's free and can do batch work.
http://www.voxengo.com/product/r8brain/

The resampler in Foobar is not of high quality (NB, I've not ABXed it, but when testing it scientifically, it performed rather poorly, even on ultra mode). Evidence http://src.infinitewave.ca/

EDIT: "not of high quality" is perhaps an exaggeration - it's just not one of the best. And if there's a better, free alternative, why no use it?

Funny, I don't see PPHS listed there.

The graphs shown are all done in 64-bit, which is far beyond the noise floor of most hardware, and definitely far beyond the noise floor of CD, to say nothing of the noise floor of vinyl. The "purple" colour is about where CD's noise floor is, so any noise that's bluer than this colour is not even going to be present in a 16-bit recording. The noise level of your vinyl is going to be around this colour, so anything that gives noise less than that should not sound different from any of the other solutions. Beware of pretty graphs; they're often misleading.

In short, foobar2000's resampler should not be audibly different from Funkstar's suggestion, especially in the case you're considering.

NB, this doesn't absolve you from TOS #8.

You're looking at the wrong graph, try selecting the cut off frequency and phase shift. No, PPHS is not listed, I performed those test myself with Adobe Audition. I agree, noise floor is irrelevant.



It does not, my apologies. I don't have the time any more to do this test justice. Perhaps someone else does?

Mind pointing out how the r8brain stuff is superior then? The free version when compared with SSRC has significantly inferior (further from ideal) graphic results in the Passband, Transition, and Phase graphs, and the Impulse graph is also representative of those inferiorities. In fact, the r8brain free resampler even inverts the phase! I hardly see, even "when testing it scientifically", how the r8brain free resampler is superior.

The only edge the r8brain resamplers appear to have is with the minimum phase resampler, which has, as anticipated, asymmetric impulse response. In fact, by all the metrics other than noise floor, SSRC is closer to ideal than r8brain.

This is not quite accurate because the noise floor on a CD is around -110 dB, considering that the spectrogram display shows a power spectral density, not the waveform level. So, the regular 16-bit dithering noise is actually represented by this color. Here is the graph of the CD noise floor with a standard TPDF dither:

Thanks for the correction. Mine was only a quick approximation.

Ah, yes you are correct. I was wrongly looking at the SSRC 0.1.2 samples which show noticeable aliasing. I'll see if I can dig out the graphs I made of PHSS as they are now where near as as clean as the SSRC.
Still, there is a SSRC plugin for Foobar, which would seem to do a great job (just check what version it is using).

I would love to see PPHS included in that list of SRC graphs you maintain. If nothing else, it's quite a useful resource. If you find them, would you mind adding them to the list?

Hi Funkstar, are you confusing SSRC with "secret rabbit code"? I'm thinking that when you refer to SSRC 0.1.2" you're really referring to the "Secret Rabbit Code" resampler. As far as I know SSRC and "Secret Rabitt" are two different things.


BTW. I've never been able to dectect any problems with foobars resampler and I've never heard/read of a single case of anyone ever being able to ABX the foobar resampler (either PPHS or SSCR).

No problem. Just run the following test signals through the SRC and send me WAV results. Only one resolution is needed, the order of preference is 32-bit int, 32-bit float, 24-bit.
http://src.infinitewave.ca/TestSignals.zip

No, I do not believe he is confusing them. The site lists the Secret Rabbit Code resampler alongside the SSRC.

Yes but one of the resamplers listed there is "Secret Rabbit Code 0.1.2 (best sinc)" so it made me suspect that when he said "SSRC 0.1.2" that he really meant Secret Rabbit Code 0.1.2.

D'oh, you are right! I have always assumed SSRC = Secret Rabbit Code. Where can I find the home page of SSRC? It seems to perform extremely well indeed, even if it does leave some aliasing.

That being said, I seem to see some discrepancies in these test. As you can see, the R8Brain Free transistion curve actually crosses the 'ideal filter' at ~-50dB. Wouldn't this cause aliasing, which would in turn be shown in the sweep graph (it isn't)?

I think that the graphs are accurate. You can see some aliasing in the Sweep test too. However it's very small (just as the Transition graph suggests), only about 200 Hz wide.

I think it's "Shibatch Sample Rate Converter" : http://shibatch.sourceforge.net/

I agree with Alexey. The graphs appear to be consistent. The fact that the transition curve crosses the ideal filter is reflected in the aliasing in the sweep graph. Notice the little "bounce" at the top? there you have it. Also, notice the slight decrease in loudness in the high frequencies?