Sample Rate Conversion: Hardware v. Software

[spektrograf]

So, I recently stumbled on this site (http://src.infinitewave.ca/) where they compare various software-based SRC implementations. The tests compare 96kHz conversion down to 44.1kHz. Clicking on "Help" brings up an interesting article on the whole subject of SRC in general.

Just by clicking through the results of a few popular DAW and audio editing apps, there are noticeable differences in the results that come out of that. That leaves me wondering...how good are the hardware-based upsampling algorithms in DAC's today? From DAC's such as the PSA DLIII's to dcs Purcell's, just how good are those hardware-based upsampling algorithms?

If someone were inclined, would they be better off doing the SRC themselves with a piece of software with a great algorithm, such as Isotope and feeding the DAC a data stream that has already been upsampled to the DAC's target upsampled sample rate... for example, 192kHz... (assuming the DAC would pass-through rather than process the data stream if the sample rate were equal from input to output) ...OR... would it be more worthwhile to let the hardware's implementation do the upsampling?

This directly applies to the DAC I have, of course... the PSA DLIII, but it also applies on the whole to high sample rate source material like HRx that is potentially downsampled on playback before hitting the digital out of a computer depending on what interface is being used.

Any thoughts on this?

[Beefy]

Any thoughts on this?

Only that it is a lot easier to change software algorithms to find one you like, rather than changing the hardware......

[deepak]

It's an area I wanted to explore more in the past. Empirical Audio recommends an upsampler used in foobar2000 that only works with foobar2000 .8.3. I never got around to trying it since the North Star sounds great with the hardware upsampling and foobar .8's GUI customization is a step backwards from .9

I've hosted the file if anyone wants to try it: here

[spektrograf]

From the "like" standpoint, I totally agree. For the price of a Purcell, anyone can try out several software implementations. Though, it would take a bit of work to convert an entire digital library over once you find one to go with. That's definitely the benefit of a good hardware implementation. If it sounds good, then it could be hugely convenient all 'round.

Thx for the link to the older version of foobar2000, deepak. I'll have to carve out some time in the near future to play around with it.

It's interesting to compare various implementations like Apple Soundtrack Pro (don't know what version was tested) to free audio apps like SoX, which offers multiple up and down sampling algorithms. STP seems to exhibit increasing noise and artifacts with respect to frequency on the sweep test. With SoX, the various algorithms seems to trade off quality vs. speed of conversion. So, purely for the sake of a technical discussion, I'm wondering what compromises are done at the hardware implementation level to achieve real time SRC at a reasonable processor cost.

[deepak]

The link is to just the ASIO dll file for foobar. I don't have the actual program, though it shouldn't be that hard to find.

I have used SoX for de-emphasizing old Japanese CDs that used pre-emphasis, and I don't think it's the most transparent method available. I think it might be adding a bit too much EQ below 16 KHz. And the imaging seems to collapse a little after running it through SoX. The only other de-emphasis EQ I've tried is the one in foobar, by manually setting the EQ. I need to try another wav editing program since I'm now using the SB.

[spektrograf]

The de-emphasis function is probably a separate algorithm altogether from the SRC algorithm within SoX. Just looking at the SRC results on that webpage, the SoX (very high quality) algorithm seems to perform superbly on all the tests

[bandpass]

I have used SoX for de-emphasizing old Japanese CDs that used pre-emphasis, and I don't think it's the most transparent method available. I think it might be adding a bit too much EQ below 16 KHz. And the imaging seems to collapse a little after running it through SoX. The only other de-emphasis EQ I've tried is the one in foobar, by manually setting the EQ. I need to try another wav editing program since I'm now using the SB.

FYI, the sox changelog says that the deemph stereo bug was fixed in sox-12.18.2 (2006-09-03).

Also the latest doc on the sox website has some words on the accuracy:

The de-emphasis filter is implemented as a biquad; its maximum deviation from the ideal response is only 0.06dB (up to 20kHz).

-bandpass