Nyquist was wrong (http://www.hydrogenaudio.org/forums/index.php?showtopic=8177&st=49&) thread (link to 1st post of Azeteg)
Test 24/96 vs CD resolution (http://www.hydrogenaudio.org/forums/index.php?showtopic=12920&hl=) thread
I've created a sample that could be useful for testing Azeteg's claims about inaudible high frequency content changing the perception of audible sounds.
I've uploaded a zip containing 3 .flac files and a text file. Here's the text:
D I S C L A I M E R
BE CAREFUL! THESE SAMPLES CONTAIN MUCH HIGH FREQUENCY CONTENT AND COULD CAUSE DAMAGE TO YOUR EARS/EQUIPMENT. PLAY BACK ON YOUR OWN RISK!
The clave_full sample is made by mixing 2 samples:
1 with content > 20kHz (clave_low) and 1 with content > 22kHz (clave_high). Clave_high is 0.8% faster, therefore high and low frequency content are synchroneous in the beginning but shifted in the end.
Samples:
clave_full_96.flac
original clava_full, 96kHz, 16bit; contains much high frequency content > 20kHz. For those who have high resolution equipment and/or want to mess with the sample themselves.
clave_full_22050lp_48.flac
clave_full, 16bit, 48kHz; lowpassed with CoolEditPro's FFT filter, 22037-22043 transition band, 24000 samples FFT size for long, strong pre-ringing. As clave_full contains tones at 22035kHz, there's quite a lot pre-ringing. This sample simulates the filtering needed for downsampling to 44.1kHz in a somehow exagerated way.
clave_low_48.flac
clave_low, 16bit, 48kHz; the high frequency content added by mixing clave_high misses.
To test Ateteg's claim try to ABX
clave_full_22050lp_48.flac
vs.
clave_low_48.flac .
If he's right, some of the clicks could sound different.
If someone needs the files in another format please tell me.
If there are other questions/remarks/suggestions please post in one of the threads linked above.
Here's the 1st 24/96 sample for 24/96 vs CD test. It is from a DVD-V containing 24/96 LPCM audio stream: Chuck Mangione - The Feeling's Back. Some more will follow.
I've tried Monkey's Audio (3.97), newest flac + wavpack from rarewares. .ape and .wavpack give similar compression ratios, flac is a little bit worse. Does anyone know a lossless encoder or settings that give better results on 24/96 material? - Thanks.
Any luck with either OptimFrog or LA? They're slower, but usually give higher compressions ratios for normal 16/44.1 Never tried 24/96 material though.
Thanks, sthayashi. I've tried optimfrog, but the results are almost identical to monkey's / wavpack (slightly better with high compression switches but awfully slow). LA homepage doesn't work properly here, so no chance to test it right now. Anyway the optimfrog result persuaded me that the compression is alright. (I wasn't sure about what ratio to expect: 24bit=more random information -> should compress worse than 16bit) BUT 96kHz=more redundancy between neighbour samples (as there's usually not much energy at high frequencies ~ 30-48kHz) -> should compress better than 44.1kHz so ... ?) I'll stick with wavpack.
BTW: 5 downloads of the 1st sample already. I hope this means (at least) 5 results will be published in the test thread (http://www.hydrogenaudio.org/forums/index.php?showtopic=12920&hl=) soon.
EDIT: Here's another sample
... next one ...
When downloading 'Mangione01.wv", I get a 1.22 MB sample. Playing it, it lasts just 4 seconds, but the player says it should be 29 seconds. The file is trimmed.
There must have been an upload problem with Mangione01.wv. Replacing it in the original post doesn't work somehow - data of the correct size is uploaded, but download doesn't work. Error message (Mozilla Firebird: "The link could not be saved. The web page might have been removed or had its name changed.")
So I'll try again in this post ...:
Edit: I've uploaded 2 times but download of mangione01_fixed.wv still doesn't work here. All other mangionexx.wv files can be downloaded. Can someone test + confirm, please?
Edit: I've uploaded 2 times but download of mangione01_fixed.wv still doesn't work here. All other mangionexx.wv files can be downloaded. Can someone test + confirm, please?
I can only download mangione02. 01 and fixed both give me an error about removed web pages.
I was able to download the screwed m01 (only 4 sec) earlier, though.
Thanks, Continuum. Seems like it's a size problem. I've replaced to not working 9MB files by two new smaller samples that can be downloaded. Here's another one (hopefully working):
Fine. Finally I've been able to upload 5 samples. The size limit seems to be close to 8 MB which makes sample lengths of ~ 20-25 seconds possible.
_______________________________
For those who think the audible difference between 24/96 and 16/44.1 (if there is any) is caused by increased (dither) noise level / (truncation) distortion, I've created a training sample: The .zip file contains 2 versions of the quietest 2 seconds of Mangione04.wv:
Training04_24.wv: amplified by +21dB -> converted to 16bit/48kHz using CoolEditPro (1bit dither, no noiseshaping)
Training04_16.wv: converted to 16bit/48kHz (same settings) -> amplified by +21dB.
By amplifying before upsampling 3.5 bits of resolution are preserved compared to doing it the other way round. So Training04_16.wv simulates the 16bit version of the file while Training04_24.wv simulates a "19.5bit version" (both at unrealistic volume). With this everyone (even without 24bit hardware) can get an impression (and maybe some training) about what the difference could sound like.
_________________________________
P.S.: It would be great if responses (unless someone wants to contribute a sample) would be posted to Test 24/96 vs CD resolution (http://www.hydrogenaudio.org/forums/index.php?showtopic=12920&hl=) thread. Thanks.
Several things:
- I can't download properly any of the samples. The wavpack samples seem to decode OK full length, but wavpack decoder reports CRC mismatches. I can't decompress the zip files either, because files are reported to be corrupted. It seems that part of the files is always cutted off.
- The wavpack samples are faded-in and out. It's not very recommendable to be so, because a tricky listener could identify dither noise at the processed files playing the beginning or the end of the samples at high volumes. Better make the samples start from a near-zero sample, but don't use fade-in our out. Fade-in or out is OK only if you apply it exactly same way at both (original and processed) test samples, but after all processing.
- After some testing, I think best tradeoff between to pre- and post-ringing and resampling artifacts is obtained using fast SSRC resampling on downsampling and slow (normal) resampling on upsampling. Fast resampling causes minimal worse performance when downsampling, and sets the ringing behaviour.
- After some analysis and quick listening tests of background noise at the samples, I think no noiseshaping (flat) dither would be ok for these samples, but to rule out chance that some true golden ears such as some of the ones at Ethan Winer 24-bit test (see ff123 page to know more) could detect the dither noise, I'd use ATH noiseshaping dither. I'd use soft ATH noiseshaping instead of strong ATH noiseshaping, because it's more than enough in this case, and because results in lower dither amplitude, including less risk of clipping. Anyway, as a side-test, flat dither could be tested to check if it makes for an audible difference.
- There's no need to use foobar2000 to perform the processing of the samples. It's way easier to use SSRC 1.29 and run a bat file that performs the whole process for each file. I'll upload a bat file that does this as soon as I learn how to do it
Ok, to speed up things (I don't have much time right now) I ve uploaded the bat files to my page.
Bat file for resampling using soft noiseshaping is at:
http://www.kikeg.arrakis.es/various/ssrc_d...own_up_dath.bat (http://www.kikeg.arrakis.es/various/ssrc_down_up_dath.bat)
Bat file for resampling using no noiseshaping (flat dither) is at:
http://www.kikeg.arrakis.es/various/ssrc_d...wn_up_dflat.bat (http://www.kikeg.arrakis.es/various/ssrc_down_up_dflat.bat)
Usage:
Copy the bat files at same location where the uncompressed wav files are. In order to use these bat files you must have SSRC 1.29 executable at this same location, or somewhere that it's at your PATH environmet variable. You must open a DOS box, go to that location (directory), and execute:
ssrc_down_up_dath <wav file without wav extension>
or
ssrc_down_up_dflat file <wav file without wav extension>
For example, if you wnat to process the mangione04.wav file, execute:
ssrc_down_up_dath mangione04
And you will end with a mangione04_up_down.wav file, result of the processing.
Edit: given the source files max. amplitude, there won't be any clipping problems due to the process.
KikeG, sorry for answering that late. I must have missed your replies.
I just tried to download some one of the .wv files (Mangione05.wv), no problems here, it's bit-identical to the one on my HDD I uploaded.
Anyway, I've created a set of PAR2 files for each sample and will upload one PAR2 file for each sample here for people that have similar problems as you have.
Usage:
1. Install Quickpar (available at sourceforge)
2. Copy the PAR file to the folder where's the .wv file and double click on the .PAR file
3. Follow the instructions; if there aren't enough valid blocks for recovery please report here and I'll upload another PAR2 file.
I'll reply to your other comments later - thanks for the effort.
PAR file for mangione04.wv
PAR file for mangione05.wv
PAR file for mangione07.wv
PAR file for mangione08.wv
- I can't download properly any of the samples. The wavpack samples seem to decode OK full length, but wavpack decoder reports CRC mismatches. I can't decompress the zip files either, because files are reported to be corrupted. It seems that part of the files is always cutted off.
I hope now everything works using the PAR2 files.
- The wavpack samples are faded-in and out. It's not very recommendable to be so, because a tricky listener could identify dither noise at the processed files playing the beginning or the end of the samples at high volumes.
Well, in real life - outside of tests like this - there are fadeins and fadeouts as well; often these fades are even much slower, so a listener has more time to focus on low volume noise than in this test. If someone really wants to cheat, this can't be prevented, no matter if fades are used or not.
Fade-in or out is OK only if you apply it exactly same way at both (original and processed) test samples, but after all processing.
I don't understand why. Of course there's a point where fading makes the original signal so quiet that dither noise becomes audible if volume is high enough (and equipment noise-less enough), but this will happen no matter what's the order of applying fade <-> processing (=resampling/changing resolution 24->16bit), won't it?
ATM I'm not persuaded yet that the problems that could be caused by the fades are big enough to justify the extra work of re-creating the samples and uploading again ...
- After some testing, I think best tradeoff between to pre- and post-ringing and resampling artifacts is obtained using fast SSRC resampling on downsampling and slow (normal) resampling on upsampling. Fast resampling causes minimal worse performance when downsampling, and sets the ringing behaviour.
From what I've seen from CEP spectral view / frequency analysis, the ringing caused by SSRC/fb2k is at > 21kHz. May I ask what makes you believe that this ringing can cause any problems so a 'tradeoff' is needed - or have I misunderstood? I don't want to question your results/recommendations but I'm curious about details ...
I'd use ATH noiseshaping dither. I'd use soft ATH noiseshaping instead of strong ATH noiseshaping, because it's more than enough in this case, and because results in lower dither amplitude, including less risk of clipping. Anyway, as a side-test, flat dither could be tested to check if it makes for an audible difference.
Fine. If you don't mind I'll edit this in the test proposal somehow.
There's no need to use foobar2000 to perform the processing of the samples. It's way easier to use SSRC 1.29 and run a bat file that performs the whole process for each file. I'll upload a bat file that does this as soon as I learn how to do it
Thanks alot for the effort! A while back (before fb2k diskwriter existed) I used SSRC for some tracks and noticed problems with too loud (clipping) signals - This was the main reason why I wanted to use fb2k for it. As you said - the samples aren't mastered at such a high volume so using SSRC is save.
If noone decides to pay more than I'm willing to at ebay I'll have got a 24/96 soundcard soon B) and can perform the test myself.
I hope now everything works using the PAR2 files.
Thanks, I didn't still have time to try them.
Well, in real life - outside of tests like this - there are fadeins and fadeouts as well; often these fades are even much slower, so a listener has more time to focus on low volume noise than in this test. If someone really wants to cheat, this can't be prevented, no matter if fades are used or not.
Yes, but, here, we are trying to evaluate if differences are perceivable just listening to the unprocessed part of the samples, at the "useful" part of them. On regular listening, nobody will pay much attention to the small faded in/out parts of a recording, and supposed superiority of 24/96 is not supposed to be just at the short faded parts of the music. However, having those fade ins and outs here can help people "cheat" at this listening test, if they just listen to those small portions of the samples at high levels. You know, given those fades, I could listen to those tiny parts at high levels for dither noise, and say I could hear a difference just with my ears, and that wouldn't be false. However, we are not interested in that. We are interested in audible differences at the unprocessed, "musical" part, where there are no fades. So I think it's a good idea just not to have those.
I don't understand why. Of course there's a point where fading makes the original signal so quiet that dither noise becomes audible if volume is high enough (and equipment noise-less enough), but this will happen no matter what's the order of applying fade <-> processing (=resampling/changing resolution 24->16bit), won't it?
No, because if you fade at the final test fines instead, fading will be in both cases made at 24 bit resolution, and the the whole signal at the resampled file will be faded, including previously added dither noise. Dither noise can become audible when it's not masked by music, that is, when it's happening over near-silence. Fading over the dithered signal won't cause this effect.
ATM I'm not persuaded yet that the problems that could be caused by the fades are big enough to justify the extra work of re-creating the samples and uploading again ...
Maybe there's no need for that. You could ask participants not to cheat, or not to listen at insane volumes, or not to focus at the fade in/out parts.
From what I've seen from CEP spectral view / frequency analysis, the ringing caused by SSRC/fb2k is at > 21kHz. May I ask what makes you believe that this ringing can cause any problems so a 'tradeoff' is needed - or have I misunderstood? I don't want to question your results/recommendations but I'm curious about details ...
I don't think this pre-ringing is likely to cause audible problems, but I think the way I suggested is more optimal. In case the pre-ringing could lead to some subtle audible differences, this way would surely be better. I prefer to take less risks.
KikeG, again thanks for the help. Hopefully I'll be able to re-do the test setup in the other thread (or start a new one), using your suggestions.
___________________________________________
Here are some more samples to get a wider mix. My thanks go to Patsoe who provided the material I took them from. The choice was hard, I created about 15 but I'll limit the ones I upload to 6 for now. If someone has d/l problems similar to KikeG, please tell me - I'll upload some PAR2 files then.
1st sample: blue_2.wv:
Blues. E-guitar, bass, percussion (sharp attacks), e-organ, male vocals
Next one ...
Desmond_1.wv
Jazz/Swing. Jazz guitar, sax, bass, percussion (some high-pitched, like cymbals)
...
Lovely_1.wv
Blues. Accoustic guitars, percussion
...
mcdougall_1.wv
Irish folk. Vocals, guitar, drums, typical instrument(s) I don't know the name of.
...
naima_1.wv
Slow Jazz. Piano, bass, percussion, horns
...
PsalmII_1.wv
Strings only
tigre,
Since you were interested in my 24bit/96kHz recordings over at Head-Fi, I thought I would upload a fresh (and cleaner) recording of myself pouring a glass of water. It is recorded using an AT 3035 LD-condensor mic into a Grace Lunatec V3 (which is a high-end 2-channel mic-preamp/ADC) at true 24bit/96kHz into CEP.
I didn't use any dynamic processing except to remove a constant noise in the background using CEP's Noise Reduction (I didn't use this on the clips over at Head-Fi). When I listend to the "Noise Only," I could barely hear the water pouring . . . so, I went ahead and applied it. If anyone would like the raw recording, I'd be happy to upload that as well.
Thanks, Douglas 256.
To save bandwith, I replace your .wav sample with a mokey's audio version.
I upload your four samples from the head-fi thread here (compressed with mokey's audio, zipped to 1 file).
I've created an oggpreview file (1.1MB) of all *music* samples I uploaded. This could help to save bandwith.
Wow... I'm surprised, but I tried abxing that sample of Water being poured into a glass, and I couldn't do it successfully. I'm using an M-Audio Transit 24/96 at the highest sampling rate settings, and I converted one sample into 44.1 using the SSRC resampler, and told the diskwriter to convert the file to 16 bit. I'm listening on some very nice Etymotic ER-4P earphones.
Of course, before testing, I turned off the resampler from DSPs. I really can't consistently tell a difference. Should I be able to, or is there only a select minority of the population that has hearing good enough to be able to tell the difference? I'm just wondering if something is wrong with my setup, perhaps. Maybe I'll try again with my M-Audio Firewire 410, which should have a bit better sound quality.
Why did you expect to hear a difference?
IIRC only one person has ever managed to positively ABX 44.1 against something higher.
Cheers,
David.
Wow... really?
Sure I expected to hear a difference. For years I've heard audiophiles lament that the quality of CDs pales in comparison to vinyl because the digital signal lacks sufficient resolution to properly convey the equivalent detail, and of course, there is excitement in some circles about DVD-A because of the superior quality of the audio.
Also, I was working on a computer at a music studio once, and I had mistakenly set the output of the soundcard to 44.1KHz, when the source material was at 48KHz. The studio engineer listened to some music for a few seconds and said "Something's wrong here..." He knew that the sampling rate was too low, but I couldn't hear a difference. (Maybe it was because resampling degraded the signal quality or something though)
I've never actually tested it for myself until now. I guess they *did* know what they were doing when they developed redbook. I imagined, at first, that I could hear greater detail in the high frequencies, but I couldn't even ABX it.
Wow... I'm surprised, but I tried abxing that sample of Water being poured into a glass, and I couldn't do it successfully. I'm using an M-Audio Transit 24/96 at the highest sampling rate settings, and I converted one sample into 44.1 using the SSRC resampler, and told the diskwriter to convert the file to 16 bit. I'm listening on some very nice Etymotic ER-4P earphones.
Of course, before testing, I turned off the resampler from DSPs. I really can't consistently tell a difference. Should I be able to, or is there only a select minority of the population that has hearing good enough to be able to tell the difference? I'm just wondering if something is wrong with my setup, perhaps. Maybe I'll try again with my M-Audio Firewire 410, which should have a bit better sound quality.
[a href="index.php?act=findpost&pid=287898"][{POST_SNAPBACK}][/a]
A bit better. Try the E-MU cards or an external DAC and a good amp. Unfortunately the ER-4P is very sensitive, so you might have issues with noise. Certainly with the Transit there must be a lot of noise.
I don't yet have an amp, but I probably will find one in the near future, just for the sake of trying things out.
The Firewire 410 was rated as the third best external audio interface over at head-fi.org
http://www4.head-fi.org/forums/showthread.php?t=112165 (http://www4.head-fi.org/forums/showthread.php?t=112165)
I'd presume it'd be good. Incidentally, I do get a small amount of noise on silence with the FW410 when the headphone output volume is maxed, but I don't hear anything on the Transit. The transit sounds really clean, and I can't hear any noise, but maybe if I compared with the FW410 I'd hear something in very quiet tracks, because the SNR *is* better on the 410.
I probably won't be buying an E-MU or Focusrite card (the two brands above the M-Audio FW410) in the near future, at least because... well... that's a lot of money for something I've already got!
The amp might be a good idea though. I'm guessing I'll have to drop about $300 for something reasonably high quality.
IIRC only one person has ever managed to positively ABX 44.1 against something higher.[a href="index.php?act=findpost&pid=288208"][{POST_SNAPBACK}][/a]
I must have missed this news.. anyone have a link?
I don't yet have an amp, but I probably will find one in the near future, just for the sake of trying things out.
The Firewire 410 was rated as the third best external audio interface over at head-fi.org
http://www4.head-fi.org/forums/showthread.php?t=112165 (http://www4.head-fi.org/forums/showthread.php?t=112165)
I'd presume it'd be good. Incidentally, I do get a small amount of noise on silence with the FW410 when the headphone output volume is maxed, but I don't hear anything on the Transit. The transit sounds really clean, and I can't hear any noise, but maybe if I compared with the FW410 I'd hear something in very quiet tracks, because the SNR *is* better on the 410.
I probably won't be buying an E-MU or Focusrite card (the two brands above the M-Audio FW410) in the near future, at least because... well... that's a lot of money for something I've already got!
The amp might be a good idea though. I'm guessing I'll have to drop about $300 for something reasonably high quality.
[a href="index.php?act=findpost&pid=288339"][{POST_SNAPBACK}][/a]
I have experience with the Transit. It's good for the money. I'm very surprised you don't notice a noise floor. If you have a desktop the E-MU 0404 is a good cheap card. That plus a maxed PIMETA plus a ER-4 P-S converter (a resistor which makes it less sensitive and makes the impedence flatter) should be within $300.
it is a 23 second 96/24 ape sample of "voy a apagar la luz/contigo aprendi" by "luis miguel". from the dvd-audio for "romances". it was recorded from the l/r audio outputs of my toshiba sd-4900 into my esi waveterminal 2496.
http://s41.yousendit.com/d.aspx?id=3OIWNJ3...RI03EUDKR93GKEI (http://s41.yousendit.com/d.aspx?id=3OIWNJ3RA8KRI03EUDKR93GKEI)
i would use the attachment feature, but the file is too big (8,176,136 bytes)