This test continues the discussion in this thread...

http://www.hydrogenaudio.org/forums/index....67619&st=75

...using the impulses provided at the end of that thread as filters.



I've filtered some 24/96 audio with the linear phase (050) and maximum phase (100) 20k impulses, and uploaded the results here:

http://rapidshare.de/files/41301790/limehouse_20k.zip.html

The "reference" is just the original convolved with an impulse, i.e. a null operation.

All files are 24-bit 96kHz sampled .wav files. You will need a 24/96 sound card to listen to these files, and use something like foobar2k which plays the files properly.

The "reference" should sound best, the "maximum phase" file should sound worst. Can anyone ABX a difference?

(I don't claim to be able to hear any difference).


btw, the files are time aligned to the nearest sample, though the maximum phase filter introduces a sub-sample delay and slight phase shift. I've confirmed that all low pass filters have identical amplitude responses, and that the pass band of the low pass filters matches the level of the unit impulse. Source: Cool Edit Pro frequency analysis, 65536 FFT, Linear View, Range 24 dB, Reference -90 dBFS, window expanded to fill my screen, absolutely no different between plots.

In the last second end of each audio file is a unit impulse, filtered - hence the filter itself is "included" in the file, for easy checking, analysis and verification.


(I used Cool Edit Pro to do the convolution (filtering). I'm not entirely convinced by its algorithm - there are some errors 100dB below the signal, verified by convolving with a single impulse. From the nature of the errors, I assume it's performing time domain convolution by frequency domain multiplication, which is quite common, and there are some rounding errors which would be lost in the noise of a 16-bit signal, but are apparent with the 24-bit signal)

Cheers,
David.

P.S. When I tried to FLAC encode these files (using FLAC 1.2.1b in Windows XP), the FLAC verifier from the same package (not the one called automatically when encoding, but the stand-alone one) reported errors, hence I've provided .wavs instead. I hope my system isn't on the way out - can anyone else reproduce this problem?

btw, the maximum phase file should sound worst.

I can't hear any difference.

Cheers,
David.

I guess this is quite a challenge then?

Cheers,
David.

I've attached some spectral plots to show how significant the ringing is at 20kHz.

limehouse1.jpg shows the original (top) and filtered (bottom).

The broad shaded vertical lines are drum hits, the little horizontal flecks at the top left of them in the bottom image are the pre-ringing of the maximum phase filter.

The vertical dotted line represents the cursor. limehouse2.jpg shows the spectrum at the cursor. You can see the spike at 20kHz. It's about 10dB less than the energy in the drum hit itself (not shown).

Cheers,
David.

I, probably like most people here, can't hear a whit above 18k to begin with. So I'm not even going to try.

According to people who believe in such things, that shouldn't disqualify you at all...

from here:
http://www.hydrogenaudio.org/forums/index....st&p=609606

Cheers,
David.

David,
I tried the reference against the maximum phase version. I decided on just 5 trials, as I find ABX listening quite a strain, but 5 trials gives a fairly convincing outcome if all answers are correct. I find a segment of about 1 or two seconds' duration that seems to sound different. I then try to avoid answering until I am quite sure of the answer.

Results were as follows:

foo_abx 1.3.3 report
foobar2000 v0.9.6
2009/01/27 19:56:49

File A: C:\Users\Public\Downloads\ha\limehouse_20k\limehouse_maximum_phase_100.wav
File B: C:\Users\Public\Downloads\ha\limehouse_20k\limehouse_reference.wav

19:56:49 : Test started.
19:59:06 : 01/01 50.0%
19:59:33 : 02/02 25.0%
20:00:29 : 03/03 12.5%
20:15:11 : 04/04 6.3%
20:16:43 : 05/05 3.1%
20:16:48 : Test finished.

----------
Total: 5/5 (3.1%)

The 'maximum phase shifted version' sounded a bit duller to my ears.

I took a long refresh break for trials 4 and 5 because my hearing had temporarily lost its ability to discriminate fine differences. Also the segment I chose to listen to for trials 4 and 5 was different from the segment I used for trials 1, 2 and 3.

I do not have particularly good high frequency hearing. These days it cuts out before 20KHz.

It is of course always possible that something in my playback equipment created the difference. There are always some lingering doubts about results like these.

Cheers

Thank you MLXXX. It's great to see someone take the time and effort to do this.


Yes, it could be something in the playback equipment, but we wouldn't get useful ABX results without any playback equipment.

What playback equipment did you use?

Cheers,
David.

Hopefully some others can do an ABX and report their results!

My playback system is a home theatre pc with an onboard high definition sound chip outputting discrete analogue channels. The analogue channels are connected to an AVR, driving medium quality hi-fi speakers.

Some details that may possibly be relevant:

• Foobar
• Vista Home Premium operating platform
• 2.4GHz core 2 duo processor in an Intel DG965WH motherboard
• Default Vista high definition audio driver (Microsoft 21/6/2006 6.0.6000.1636)
• Analogue connections from motherboard onboard high definition audio to Yamaha AVR (model HTR-5750). Tone controls defeated.
• Magnat Vintage 350 main speakers
• Listening volume: comfortable, not particularly loud
• Listening distance: 3 metres