Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: ABX Just Destroyed My Ego (Read 100196 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

ABX Just Destroyed My Ego

Reply #50 –
Quote from: Pio2001 on
sinc interpolation  ? Given the awful results that it gives for picture, I don't think that this is the best way to resample ! It causes a tremendous amount of ringing at the cutoff frequency. 22050 Hz is usually inaudible, but in case of non-linear behaviour from an element of the hifi system, it is safer to use a more classic filter.
The standard filters for CD, for example, go from 0 dB at 20 kHz to -140 dB at 22049 Hz (not sure about the -140, but it is something well below -96 dB). Sinc go from 0 dB at 22049 Hz to minus infinite at 22050 Hz.


I don't know anything about resampling  Maybe I should read about different resamplers (PPSH, SSRC, SRC) and their settings.. from somewhere.

ABX Just Destroyed My Ego

Reply #51 –
Quote from: John Stimson on
Quote from: Mike Giacomelli on
Just a heads up, but when you ABX, you need to decide how many trials you'll do in advance.  If you stop when you're ahead, you defeat the purpose of the test.
That really depends on the purpose of your test.  If the purpose is to create statistically based scientific evidence, then you are correct.  You cannot prove anything by using the "continue until you succeed" strategy.  That is similar to doubling your bet at the casino until you win -- while it works in theory (but not in practice -- even Bill Gates isn't rich enough to keep doubling indefinitely) that fact does not prove that the odds on an individual bet are in your favor.

However, an ABX tool can also be used to gauge for yourself whether you can hear the difference between two tracks.  In that case, you are trusting your own judgement about what you can hear, and the cumulative results are simply one more piece of information you can take into account.  In that case, you are not performing a true ABX experiment.  You are simply using an ABX tool to assist yourself in comparing tracks.



Because he posted the results, I think its obvious the purpose of his test was to see if he could ABX it.  You're right, if you don't care about the results, you can do whatever you like, but thats not really what I was talking about.

ABX Just Destroyed My Ego

Reply #52 –
Quote from: Mike Giacomelli on
Just a heads up, but when you ABX, you need to decide how many trials you'll do in advance.  If you stop when you're ahead, you defeat the purpose of the test.


I know that stopping at 8/8 doesn't prove I "passed" the test, but I wanted to point out that I didn't pass, and for that purpose I thought 10/13 were sufficient.

Anyway - since I still have the encodings on my HD, I did a second test today, going to 16 no-matter-what-happens, and it ended 11/16. No surprise for me.

Though I have to confess that considering the 10/13 result, the comments I made about the warbling and lowpass in the same line are obsolete, as I couldn't distinguish the Q 4 sample anymore. Sorry about misleading you.
Nothing is impossible if you don't need to do it yourself.

ABX Just Destroyed My Ego

Reply #53 –
Well this thread inspired me to test myself... I successfully ABX'd fatboy, castanets and the harpsicord samples using Lame and Nero AAC thusly:

Harpsicord:
Lame -V2 --vbr-new
Nero -q 0.4

Fatboy:
Lame -V7 --vbr-new
Nero -q 0.6

Castanets:
Lame -V9 --vbr-new
Nero -q 0.3

First thing to note is that on Harpsicord, and no previous training, I managed to defeat --preset fast standard! That was suprising, actually (figured I would be a poser).

On fatboy I defeated Nero's standard profile (0.5), and the one above it as well. Lame shut me down hard, which is surprising given MP3's supposedly inferior low-bitrate encoding.

Castanets I just had trouble with... I know what to listen for, but it just gets transparent too easily on both encoders.

This was fun, and I'll probably do it again some time! For now, it looks like Nero -q 0.5 should be satisfactory for my iPod use (I don't listen to a lot of Fatboy ).

ABX Just Destroyed My Ego

Reply #54 –
My ABX tests were the one reason I started ripping to -V 4 --vbr-new, Nero AAC at -q 0.35, and iTunes AAC at 128kbps VBR (depending on what I feal like).  I conducted a ABX test with my equipment and I could somewhat spot -V 5 --vbr-new (it was hit or miss with some samples but nothing to get in a bunch over).

I stopped fooling myself and now have a lower bitrate library.  Funny how ABX testing can do this.

ABX Just Destroyed My Ego

Reply #55 –
I tried the same Depeche Mode track with Vorbis (aoTuV b4.51) and Musepack (1.15v).

I managed to ABX successfully musepack thumb preset (86 kbps) and radio preset (126 kbps), and for the first time of my life, Vorbis at q4 ! (126 kbps) 

Funny how codecs sounds. Mp3 sounds Skrotchgloosploof, Musepack sounds kwishwikwilwi, and Vorbis just shhhh !

ABX Just Destroyed My Ego

Reply #56 –
Quote from: Pio2001 on
----snip----
Funny how codecs sounds. Mp3 sounds Skrotchgloosploof, Musepack sounds kwishwikwilwi, and Vorbis just shhhh !


eh? 
Reason is immortal, all else mortal
- Pythagoras

ABX Just Destroyed My Ego

Reply #57 –
My experience with ABX-ing (and listening tests in general for that matter) is that the equipment, the music, and the environment make all the difference.

In a quite room on a pair of studio monitors, I could easily ABX .wav from .mp3 at all levels (classical piano sample... I could tell anything below -V2 when it was rock music, -V0 and -V1 were transparent to my ears).

On my laptop, in a coffee house, using Sennheiser PX100's (which how I listend to music 75-80% of the time) I couldn't ABX the same .wav from -V5 or 128 AAC VBR (iTunes).

As has been mentioned before and I'm sure above... know where, how and what you listen to and encode accordingly.

ABX Just Destroyed My Ego

Reply #58 –
Thanks to my Sennheiser PX100's I can ABX Lame -V5, Vorbis -q4 and iTunes 128kbps VBR, from the original file

I still use -q4 Vorbis for my iPod, I'll compromise a tiny loss in quality so I can fit more music on it

ABX Just Destroyed My Ego

Reply #59 –
In general I tend to not believe claims of successful abx tests unless actual logs are posted -- exceptions being for known killer samples, bitrates lower than 128, or from members well-known for their abxing contributions.

ABX Just Destroyed My Ego

Reply #61 –
This thread and the latest multi-format test has caused me to change my library to 128 itunes vbr. I cannot tell the difference between lossless and 128 vbr aac itunes. I was born with a slight deafness. I'm not hard of hearing but I can't hear certain frequencies. I remember taking audio tests in school and I'd be embarrassed because I wouldn't hear certain tones. Sometimes I'd just press the button in certain intervals just in case.

I'm kind of bummed I can't tell the difference because I feel like I'm missing out on something as great as lossless audio. I have AKG K701s with a Total BitHead amp and that's helped me bring me closer to experiencing better audio but to me 128 kbps vbr sounds really good and I can hear "all" the details just the same as the lossless.

Maybe my ear is untrained but I'm guessing its mostly my slight deafness.

ABX Just Destroyed My Ego

Reply #62 –
Quote from: volcomjerk on
I'm kind of bummed I can't tell the difference because I feel like I'm missing out on something as great as lossless audio. I have AKG K701s with a Total BitHead amp and that's helped me bring me closer to experiencing better audio but to me 128 kbps vbr sounds really good and I can hear "all" the details just the same as the lossless.



If you are enjoying your music at 128 itunes vbr (i'm guessing you use the AAC encoder) then don't be bummed, just enjoy it.  It's better than obsessing on what kind of encoding you should be using.

ABX Just Destroyed My Ego

Reply #63 –
Maybe i'm listening to the wrong kind of music to hear more things. Can anyone suggest something more complex? I listen to a lot of rock and punk and I guess things get drowned out in the distortion of the guitars.

I guess I'm not obsessing over what encoding to use, I guess I'm just worried that I'm not hearing what I'm suppose to hear.

ABX Just Destroyed My Ego

Reply #64 –
Quote from: volcomjerk on
Maybe i'm listening to the wrong kind of music to hear more things. Can anyone suggest something more complex? I listen to a lot of rock and punk and I guess things get drowned out in the distortion of the guitars.

I guess I'm not obsessing over what encoding to use, I guess I'm just worried that I'm not hearing what I'm suppose to hear.

Trumpets.  Or harpsichords.  Both are killers.

ABX Just Destroyed My Ego

Reply #65 –
Quote from: Pio2001 on
sinc interpolation  ? Given the awful results that it gives for picture, I don't think that this is the best way to resample ! It causes a tremendous amount of ringing at the cutoff frequency. 22050 Hz is usually inaudible, but in case of non-linear behaviour from an element of the hifi system, it is safer to use a more classic filter.
The standard filters for CD, for example, go from 0 dB at 20 kHz to -140 dB at 22049 Hz (not sure about the -140, but it is something well below -96 dB). Sinc go from 0 dB at 22049 Hz to minus infinite at 22050 Hz.


Sorry for answering to such an old post, and OT as well :-)
I'm currently reading up on a few audio technology topics to better understand what's going on (and maybe to contribute to some audio project later). So I'm asking just to make sure I understood correctly. Shouldn't sinc interpolation give (ideally) perfect results when upsampling? As I understood it, when you only add the sinc values up to a certain distance from the sample you're calculating, you get inaccuracies of course, but these should only be noticable when there are frequencies close to half the sampling rate of the original signal, since the contribution from lower frequencies mostly cancels out.

Did I misunderstand something there?

ABX Just Destroyed My Ego

Reply #66 –
The sinc interpolation gives a perfect brickwall. Try it at 11 kHz if you can and see by yourself. It introduces a lot of ringing. A perfect brickwall is usually not wanted.

ABX Just Destroyed My Ego

Reply #67 –
Quote from: Pio2001 on
The sinc interpolation gives a perfect brickwall. Try it at 11 kHz if you can and see by yourself. It introduces a lot of ringing. A perfect brickwall is usually not wanted.


What do you mean by "try it at 11 khz"? Upsampling from 11 khz? upsampling to 11 khz? I was only referring to upsampling in my last post, of course downsampling without using a lowpass before causes aliasing.

ABX Just Destroyed My Ego

Reply #68 –
The sinc is not an upsampling or downsampling algorithm. It is a lowpass filter. Also called antialias filter when used together with upsampling or downsampling. I suggested to remove frequencies above 11 kHz using a sinc lowpass.

In ff123's sample page, the Mustang samples, aimed at testing low-pass audibility, used to be filtered with sinc. I complained that the ringing was so obvious that it could be ABXed separately from the low pass itself. He then replaced the samples by new ones, filtered more progressively. They now sound as lacking treble, with no audible artifacts.

ABX Just Destroyed My Ego

Reply #69 –
Quote from: Pio2001 on
The sinc is not an upsampling or downsampling algorithm. It is a lowpass filter. Also called antialias filter when used together with upsampling or downsampling. I suggested to remove frequencies above 11 kHz using a sinc lowpass.

In ff123's sample page, the Mustang samples, aimed at testing low-pass audibility, used to be filtered with sinc. I complained that the ringing was so obvious that it could be ABXed separately from the low pass itself. He then replaced the samples by new ones, filtered more progressively. They now sound as lacking treble, with no audible artifacts.


Thanks for your replies, I learn a lot by being contradicted. I did not know the sinc function could also be used for lowpass filtering. My understanding is the following: When you sample a function, you can reconstruct the original perfectly as long as the function is sampled with more than twice its highest frequency component (Nyquist's sampling theorem). I read that the way to do this reconstruction was by using sinc interpolation (Whittaker-Shannon interpolation formula).
So if you sample this reconstructed function with a higher rate, you should have a perfectly upsampled signal.
Where does the ringing come in?

ABX Just Destroyed My Ego

Reply #70 –
Quote from: MedO on
Where does the ringing come in?

It's in the time domain. A sinc filter will transform an impulse (think one sample at 1 surrounded by 0:s) into a sinc shape. So you will spread the energy from being infinitely narrow to be infinitely wide - sure, after the impulse the ringing is probably masked, but you will have something similar to pre-echo before the impulse which can be easily heard.

ABX Just Destroyed My Ego

Reply #71 –
Quote from: ErikS on
Quote from: MedO on

Where does the ringing come in?

It's in the time domain. A sinc filter will transform an impulse (think one sample at 1 surrounded by 0:s) into a sinc shape. So you will spread the energy from being infinitely narrow to be infinitely wide - sure, after the impulse the ringing is probably masked, but you will have something similar to pre-echo before the impulse which can be easily heard.


I see your point. A single pulse like you describe would, after upsampling, be heard as a signal of half the original sampling frequency. But I still think that this interpolation is correct. An impulse of no width is not a bandlimited signal. I'm always assuming that the original signal has been filtered before recording to exclude any frequency above half the sampling frequency. If the sampled signal then looks like 000000010000000, then the filtered input signal did have a sinc shape.

edit: Maybe sinc gives bad results because actual recording equipment cannot lowpass an input signal ideally?

ABX Just Destroyed My Ego

Reply #72 –
Quote from: MedO on
So if you sample this reconstructed function with a higher rate, you should have a perfectly upsampled signal.
Where does the ringing come in?


It comes from the resctrictions needed to apply the sampling theorem : a given sample rate allows you to reproduce frequencies inferior to half the sample rate. In real life, analog signals are not filtered in such a mathematical way. They have a small amount of energy outside their main frequency range.
When you deal with digital, you assume that your signal has a strictly null amplitude above half the sampling frequency. That's your "perfect signal". It is not the original one. It is the original one minus the peanuts above half the sample rate. Taking back these peanuts introduces ringing.

If you work at 44100 Hz, and your signal have significant content near 22050 Hz, you define a "perfect signal" as having a brutal discontinuity in frequency response. There is something near 21 kHz, then, absolute zero at 22050 Hz exactly. This discontinuity in the frequency domain equals a lot of 22050 Hz oscillations in the time domain.

They are not a problem, since 22.05 kHz is usually an inaudible frequency. But if you do the same at 11 kHz, then you will get a disturbing 11 kHz resonance in addition to your music. This is not an inaccuracy or imperfection of the process. This is a signal with a brutal discontinuity in frequency response. This is how it is supposed to sound.
Cutting frequencies above a given threshold is required for the sampling theorem to work. In practice, we don't want to cut them suddenly, the resulting signal is not very interesting. We prefer recording signals with softly limited frequency response.

ABX Just Destroyed My Ego

Reply #73 –
Given the excellent results of Vorbis in recent tests, and given that Vorbis was the only codec that I could not ABX at all in Sebastian Mares 128 kbps multiformat test, I consider switching from Musepack to Vorbis.

I have just tested Vorbis 2.83 aotuv beta 4.51 -q5 with my collection of killer samples. Here are the results :

Amnesia : transparent
Astral : transparent
Drone : transparent
Fsol : transparent
Ravebase : transparent
Rebecca : transparent (not a killer sample)
Short : transparent
Spahm : ABX 16/16. Not very good.
Transwave : ABX 7/8, then I decided to go on to 24. Total 23/24. Some slight artifacts.
Vilbel : 16/16, awful !

I tried vilbel at -q6, same thing. ABX 16/16, awful.
At q7, it is transparent for me.

I'm currently trying Transwave and Spahm at -q6, but its quite hard.

ABX Just Destroyed My Ego

Reply #74 –
I have some serious hearing problems.. From birth to age 10 or so, I had tubes in my ears, was partially deaf, and they'd always puss... Disgusting..

I'm 20 now, and I still have to turn the TV way up to hear it.. The interesting thing is that I can pick out artifacts very easily. I had no formal training with ABX testing, had some cheap $20 sony headphones, and ABX'd 128kbps AAC+ 10/10 (I think, I'll post the logs if need be... I have them saved on my server.. And it wasn't a formal test with perticular encoders, it was some quick test using Winamp's encoder... This isn't the point I'm trying to make though).. The point, however, is that audio codecs are subjective. Every encoder has its own method of hiding its shortcomings. So you think your ears suck? Honestly, it's highly doubtful. Lots of people can be trained to hear whats wrong with audio encoding schemes.. Some people are pickier than others.

Audio encoders are perceptual (obviously). Most people, when they are looking for flaws, they'll find them. If you know what to look for, it's a lot easier than blindly looking for problems. While I don't have golden ears, I'm good at finding something and grabbing onto it, which is the only reason I believe I passed the ABX test I mentioned above so well. My hearing sucks horribly, probably worse than most people here. However, I believe that audible tests are more mental than physical, and it can be proven by my TV volume that almost always sits at 20 (out of 34), and my constant "What??" over and overs.

When people tell you you need training to ABX, they are absolutely right. I believe almost ANYBODY can ABX something if they know what to look for. While I didn't have formal training, I did a lot of experimenting to figure out what to look for. My brain tells me whats wrong, not my ears. If you can't do 64kbps, then do some experimenting (actually, you know what? Don't. It'll ruin music for you... Ignorance is bliss when you can fit 2x-3x more music than someone who knows what to look for... That's one thing I miss... When downloading a 64kbps mp3 simply meant 'faster download' to me, I was much happier)

There is an upside and a downside to everything. Golden ears are good when I want to help my friend (he writes a media player) test new psychoacoustic tunings in the decoder, I'm happy that I know what I'm doing to some extent. Do I wish that I was one of the people who had no problem downloading 96kbps mp3 files? Yeah, sometimes I do.

I'm glad that his media player got good reviews for sound quality, but I'm mad that I can't fit 20,000 songs instead of 8,000 on my harddrive.

To end this long post, if you want to ABX, you need to be aware of the fact that although it may be fun for a second to brag about how you can hear certain things that others can't, you can say goodbye to being blissfully ignorant of the little things, like being able to play a song without having to 'approve' it before it goes into your media library. I lurk HA all the time, and the original poster struck a chord with me... Some may think its great to hear a bunch of things in audio, but I can honestly say I got bored with it very quickly. I'll go back to lurking now, since I'm apparently in a 'downer' mood right now... But I did feel that I needed to state my true opinions on sound quality... And when I get my setup fixed (damaged due to flooding last week), I'll probably be happy again... I guess I'm just mad that I didn't make lossless backups of my music that I lost in the flood.. When I post something else, pretend this post didn't happen. Thanks.
q4 AoTuV Vorbis is my friend.