Bye Bye AC'97!!!!!! Huzzah!

[CosmoKramer]

Real question is what speakers or headphones were you using? I can tell pretty easily with the Tannoy monitors or Sennheiser HD570s I use regularly.

Show me some ABX (double blind test) logs and I might believe you. .

My equipment is a pair of HD580s hooked into an M-Audio Revolution.

 

[Neeyik]

I'm surprised you didn't note a difference between 16 and 24-bit sound though.

 

[CosmoKramer]

I'm surprised you didn't note a difference between 16 and 24-bit sound though.

Given that when noiseshaped dither is used properly 16 bits has an audible dynamic range of 110+ dB, what would be so surprising about that? 24-bits definately has its place in recording but it is (arguably) not needed as a delivery format.

For those having the proper equipment (nonresampling soundcard capable of at least 24/96 playback) here is a page with test samples (go to the bottom of the page): http://64.41.69.21/technical/sample_rates/index.htm

In order to perform a double blind test I suggest downloading the excellent audio player Foobar2000 at http://www.foobar2000.org/. There is an ABX plugin available here: http://www.foobar2000.org/foo_abx.zip

 

[pcchen]

I tried the first sample (Castanets-2) in 24/96 format. I used an Echo Indigo card (a PCMCIA sound card) and a Sony E-888 earbud. The 22/16 and 28/16 samples has apparent differences (22/16 is very clear even on low volume, but I need to set volume higher to hear the difference in 28/16), but I can't tell the difference in other samples. I have no golden ear, you know

 

[DemoCoder]

Bah. The more important thing is support for 5.1+ channels. You can definately tell the difference there.

 

[CosmoKramer]

I tried the first sample (Castanets-2) in 24/96 format. I used an Echo Indigo card (a PCMCIA sound card) and a Sony E-888 earbud. The 22/16 and 28/16 samples has apparent differences (22/16 is very clear even on low volume, but I need to set volume higher to hear the difference in 28/16),

1. Congratulations
2. Were those tests double blind? How many trials did you do? What was your success ratio? I'm afraid if you didn't perform ABX tests what you wrote above is worthless. Placebo is very powerful.

Edit: For this discussion you should only compare the reference files, i e the 16/44.1 reference vs the 24/96 reference.

Bah. The more important thing is support for 5.1+ channels. You can definately tell the difference there.

Absolutely.

 

[pcchen]

1. Congratulations
2. Were those tests double blind? How many trials did you do? What was your success ratio? I'm afraid if you didn't perform ABX tests what you wrote above is worthless. Placebo is very powerful.

Edit: For this discussion you should only compare the reference files, i e the 16/44.1 reference vs the 24/96 reference.

According to the website, one should use all test files in the same sample rates because some soundcards have a slight delay in sample rate change. So if you click "Play A" then "Play X", if there's a slight delay, you know they are different sample rate. However, it appears that the downsample operation is not perfect. I do the tests up to 20 times (only one or two times wrong guess) where the foobar's abx plug-in failed to give a probablity number.

I tried to use the two wave files with different sample rates, but unfortunately there is a slight delay between sample rate change on my card, therefore it has no point to do ABX on these. But the apparent artifacts are not that clear in a genuine 44 kHz file (than the downsampled 96/24 file), I'd say.

By the way, the RMAA test shows my card performs worse when in 96 kHz mode. However, such difference is not likely to be audible by normal human (about 0.4 dB difference at 15 kHz, I probably can't hear anything near 15 kHz though).

 

[CosmoKramer]

According to the website, one should use all test files in the same sample rates because some soundcards have a slight delay in sample rate change.

This can be circumvented if you let each test sample play through its entire length. If you want to isolate a specific passage the foobar abx plugin lets you manually choose any interval.

I do the tests up to 20 times (only one or two times wrong guess)

Using what application?

where the foobar's abx plug-in failed to give a probablity number.

What?? The Foobar plugin always provides a probability number.

But the apparent artifacts are not that clear in a genuine 44 kHz file (than the downsampled 96/24 file), I'd say.

Yes, some people are very sensitive to various downsampling algorithms.

 

[pcchen]

This can be circumvented if you let each test sample play through its entire length. If you want to isolate a specific passage the foobar abx plugin lets you manually choose any interval.

Not work very well since I can hear a little pop noise after sample rate change...

What?? The Foobar plugin always provides a probability number.

Not after 20 times... the probability number became NA.

 

[CosmoKramer]

Not after 20 times... the probability number became NA.

Hm...weird.