I did some test using RightMark Audio Analyzer, with asyncronous testing. I played the generated wav file in foobar v0.5 using different output modes, and I recorded it using Sound Forge 6.0. The loopback was a software loopback, using the DMX6Fire control panel / record source: Mix option. Luckily the kernel streaming result was bit identical to the reference results (using generated file as recorded file did the same values as the KS-ed file). So I like my card and foobar.

I did some other tests using some different modes, and the results were much worse. You can find these results here. I think only my results are not enough for general discussion so download the software, select asyncronous mode, mute every input source and test your card. I'm almost sure that you can't find a card, what is able to do almost perfect test results in DirectSound/Waveout mode. You can do some CD player test too, by burning the generated wav to the cd and record it using digital input, but I don't have a cd player with a digital out.

Try to check an Audigy with a 44.1 kHz file (with EAX and Sensura3d enabled too, using Creative DirectSound drivers.. ).


Pancserzso

awesome stuff, could you experiment on upsampling to 48kHz and 96kHz? and maybe the difference between 24bit and 16bit.

1) Audigy 1 does not support 96kHz or 24 bit output.

2) No Creative cards use Sensaura, as they already support EAX and A3D natively. Not to say Sensaura is software-only, of course...

Interesting, it seems that in your case something is changing the sound streams, probably kmixer. I did a similar test with my Audiophile 2496 but the output seemed identical in all cases. However, I'll try this same test to see what happens.

Did you restart foobar between changing playing modes? (Because it is needed to change, exit and start the program to change the way of playing)

I did tests both with Winamp and Foobar. Anyway, the Foobar version I used seemed to change output mode without need to restart.

I have done some more test about bit depths and the effect of dithering: I played a 32 bit 44.1 kHz wave file in foobar using KS mode and recorded it in Sound Forge using 32 bit 44.1 kHz option. Only the bit depth and the dithering was changed during the tests.

You can see the results showing that 32 bit non dithered is the closest to the reference (i don't know how can it be better in S/N part than the reference ) but the others are not that good. (The test could only output 4 chart at the same time, so I didn't included the 32 bit dithered, but it was a little bit better than the 24 bit dithered, but much worse than the 32 bit non dithered.) I've made some resampling test (you asked...) You can see the ugly results here.

So the conclusion:
Use the 32 bit non dithered mode, if your card supports it, if not, then the 24 bit dithered. (And buy a new card if it only supports 16 bit 44.1 khz...)

And if you can afford it (you are an owner of a non-resampling card) do not use any resampler!

Can you publish somehow your results for the 44.1 kHz 16 bit file? (Just pust the save as HTML button and put it to a free provider.)

I would't extract many conclusions, here the limiting factor seems to be kmixer changing the streams, and according to my experience that depends a lot on your souncardcard driver implementation. Also, the results aren't coherent with what should be, so I guess that your card drivers respond in different ways depending on the bitdepth fed. Maybe the recording procedure (recording from internal mixer) behaves in a different manner depending on the bitdepth, I'd say it is not a very reliable method. Also, this method will surely work in a different way in other cards.

A better procedure would be to record the digital output of your card, or a high quality recording of the analog output, using ks output plugin. For comparison, you could try also waveout and directsound. I'll try it when I have some time, and post results.