2013/4/3 Peder Hedlund <peder(a)musikhuset.org>rg>: > In view of the recent debate regarding the alledged "crappiness" of MP3 I > thought it would be fun to see if the LAU society can tell lame (v3.99.4) > MP3s from the original.

trials 10 against 165: correct 5, p-value 0.623 against 124: correct 7, p-value 0.172 against 108: correct 9, p-value 0.011

side effects: back to the '80 and doing the moonwalk B-)

2013/4/3 Peder Hedlund <peder(a)musikhuset.org>rg>: Hi, trials 10 against 165: correct 5, p-value 0.623 against 124: correct 7, p-value 0.172 against 108: correct 9, p-value 0.011 side effects: back to the '80 and doing the moonwalk B-)

Quoting Ralf Mardorf <ralf.mardorf(a)alice-dsl.net>et>: Since Raffaele was able to tell the difference with the 108kbps file, was pretty successful with the 124 and failed with the 165 I'd say the example is definitely useful, since that's the kind of numbers you'd expect.

On Wed, 2013-04-03 at 13:54 +0200, Raffaele Morelli wrote: > 2013/4/3 Peder Hedlund <peder(a)musikhuset.org>rg>: > > In view of the recent debate regarding the alledged "crappiness" of

> > thought it would be fun to see if the LAU society can tell lame

> > MP3s from the original. > > > > Everyone is invited to download the testfiles from > > http://www.musikhuset.org/~peder/AxelF.zip , see if you can ABX them

> > post the result. > > There's one original WAV file and then 3 MP3s of various bitrates,

Playing one file after the other using VLC -> Jack @ 48KHz -> HDSPe AIO phones out -> AKG K 240 DF I didn't notice a difference for the audio quality. _But_ the mix of the original wav for sure is nice for that kind of music, but it isn't a high quality audio recording useful for this test.

So here is the thing, it makes a large difference what the source is as to how obvious it would be. Case in point, I found the v6 difficult on the files in the thread, but I toss in one of my recordings(Using lame with v4 and v6 as above) from a live concert and holy crap is there a difference, obvious right from the get go in the hiss in the track on v6. v4 on the other hand was very difficult to tell on that track to where I don't think I could hear a difference. But that only makes me wish I had some decent recording of full orchestral classical at my workstation to listen to as that is my goto material for a source. Or irish folk music, which can do a decent job as well causing problems of more than one sort:) Some things to keep in mind, I am listening over studio monitors, not headphones, but still much better than the average listening environment. It would likely be even more obvious over headphones to tell the truth. Second it always makes a difference in the second generation encoding, which is why I tend to use FLAC for archival and master purposes, and use OGG or otherwise for listening, so that I never risk a second generation compression, which becomes more obvious. Via abx, there wasn't ever a question with v8 really. v6 as I mentioned on the sources mentioned above were much harder, but if I tossed in my own acoustic/folk recording it became very blatant. Again source makes all the difference. v4 I couldn't tell over speakers, but I also don't have most of my music collection here either, so I really should come back to it with some decent classical and irish folk tracks and see what happens. Seablade

Agreed the source material probably makes quite a difference. I'd imagine there's music that would be transparent at ~100 kbps while other might require ~160. As previously stated I do believe the AxelF sample is in the harder-to-encode region for MP3 encoders so I'm pretty sure my concentrated listening threshold for average music is somewhere around 130-140 kbps or so.

On Fri, Apr 5, 2013 at 7:44 AM, Thomas Vecchione <seablaede(a)gmail.com>wrote;wrote: this isn't really true. MP3 comes out of generic psycho-acoustic research and isn't based on the input material at all - it is based on the physiology and psychology of human hearing. no, that isn't why. there are reasons, but its not because of what "MP3 aimed at".

Quoting Thomas Vecchione <seablaede(a)gmail.com>om>: Remember the 80s pop synth sound you have in that is right where AFAIK MP3 was developed to make Tom's Diner by Suzan Vega sound OK :) As seen in my test results the non-transparency in V6 (aka 124 kbps) was pretty obvious. It might very well be equally obvious using your live recording (please post 10s of it somewhere and I'll check it out) but since this particular snippet of Axel F fails for me with V6, I doubt it's an "easy to encode" file. - Peder

As previously stated I do believe the AxelF sample is in the harder-to-encode region for MP3 encoders so I'm pretty sure my concentrated listening threshold for average music is somewhere around 130-140 kbps or so. Actually I really don't think it is. The material I mentioned will probably be much harder on it. And especially non-digitally recorded or mastered material, ie things with what would normally be considered 'noise'.

Don't we have fair use laws in most of the world? Academic research (testing encoders) should be covered by it, no?

first off all, synth-pop could be a good source too, there are a lot of Peter Gabrial recordings, that are better recorded than most classic recordings, but for sure Axel F. is absolutely not useful for this test.

Peder I listened to the Axel F. samples one after the other without "dissecting" the sound. I didn't hear a difference, that doesn't mean I wouldn't hear a difference by an ABX test or by listening to it focused and for two or three times instead just for one time.

However, if there isn't really significant loss, it doesn't matter, if there is or if there isn't minimal loss, since the music anyway will sound different on different gear.

For my individual dealing with audio recordings I can't imagine why I ever should use MP3 or any other codec.

I don't need data reduction, if I listen to audio recordings, than I do it "conscious", I never ever will take a walk and listen to music.

On Sat, 2013-04-06 at 01:51 +0200, Peder Hedlund wrote: The good sound quality is the transparency of the sound, the room, the dynamic. IOW a loudness war mix takes away all the quality of the original Beatles recordings and IMO codecs with so called inaudible loss do it in a similar way.

I didn't hear a difference,

Again and again and again, that MP3 is good enough for my taste to preserve a few seconds of Axel F., is evidence for nothing. ABX tests should be made with transparent recordings of a 40 instrument orchestra or similar. I don't say that Axel F. is bad music, bad I say it's very simple synth music, were not that much can go wrong by a little bit of loss.

In taberna quando sumus, we could first drink on the libre to be free to like or not to like lossy codecs. If the codec is ok for the tests, you don't have a guaranty that it will be ok for another recording at another time, so for me there's no reason to risk loss, resp. always to test listen if the MP3 is ok or not ok for a specific recording. Use MP3! I don't use it!

MP3 isn't just a thorn in my side regarding to it's quality, but also to the way music is misused today. MP3 for me stands for music 24/7. I don't eat chocolate the whole day, I don't have sex the whole day and I even won't hear good music the whole day.

Thanks for posting this. I cannot tell the difference with ABX - even for the 108 version - so I guess either my ears are shot (quite likely) or my ears are not able to hear such subtleties (also quire likely)..

On Fri, Apr 05, 2013 at 11:42:31PM +0200, Ralf Mardorf wrote: One of the more interesting jobs I've ever done was to define a procedure to select 4 matching microphone capsules for Ambisonic mics. That's a lot more difficult than finding good stereo pairs, and you can forget about doing it by listening. The only thing you could probably achieve that way is to identify a capsule as completely out of spec and unusable. Nor can you do it by *simple* tests, which is maybe the only alternative to listening you can imagine. It requires, amongst a lot of other things, comparing directivity factors and phase response over the entire frequency range, using criteria that are matched to the question to be answered : can these four mics be used together for an AMB mic or not. It excludes factors that would easily lead you astray when listening but which are in the end irrelevant.

Quoting Ralf Mardorf <ralf.mardorf(a)alice-dsl.net>et>: No, data reduction in MP3, Vorbis, Opus, AAC or any of the other technologies is not just like a regular compressor in any way, shape or form. Neither in the way of a 3:1, soft knee@-18 dB compressor or in the way of a "repeat this 0101 pattern 24 times" zip file.

It's exploiting the psychoacoustical fact that the sensitivity of our ears depends a lot on which frequency we're listening to and that a strong signal in one frequency completely masks a weak signal nearby to our ears. So if you have a -20 dB signal at 7kHz at the same time you have a -5dB signal at 7.2kHz you can discard the former one since we won't hear it anyway.

There is an extension to MP3 called MP3 surround which is able to handle 5.1 surround sound (2D and 3D are image formats, not audio formats).

Many people nowadays aren't aware how to make sound transparent even with cheap analog home recording gear, but it's possible. We even don't need a compressor, many things a compressor does, can be done by EQs.

I try to find better words in English now. I care less about the bandwidth of the frequency. If the bandwith becomes smaller or does colour some frequencies, it's ok for my taste, assumed it is not too much. If the range endet at 18KHz for the original and a copy or transformed by a codec has loss, by ending at 15KHz, I'm able to notice this, but won't name it "I hear loss of highs", as long as the transparency and loudness dynamic, IOW the atmosphere of the recording is the same. It's important that the loudness ratio between the instruments is kept, the room is kept. Good recordings made with a 4-Track cassette recorder do sound less good than a recording with professional gear, if you take care about those things, when listening on professional gear, but it's impossible to hear it on an averaged consumer hifi. Taking care about the bandwith of the frequencies is not completely unimportant, but even something important as signal separation, is not very important, as long as you take care about this limitations, when doing the mastering. The music will sound completely different biased on different consumer hifi gear, it doesn't matter if it was recorded in a professional or in a home studio. A good audio engineer takes care that the sound is most transparent, even if the frequencies are cut etc.. Sure, if you play a good recording on a very bad hifi, it does sound ugly, but very often it does sound better, than at least the old MP3s I know, because the ratio between the instruments etc. seems to be more biased by MP3s than by bad gear. Optimal is 48 KHz, but I also mentioned that for some recordings even 32 KHz is ok. For me it's bad when "transparent" sound becomes "muddy". "Muddy" doesn't mean less highs only (it often comes with less highs), "muddy" is for a loss of transparency. The original sound might separate guitar and trompete very good, even when they are in the same position of the room, but they have accentuations by different frequency bands. If a codec does remove frequencies, those instruments will be less good separated. This is the kind of loss I can't tolerate. The loss of highs and a little bit noise are less important for me. Many people nowadays aren't aware how to make sound transparent even with cheap analog home recording gear, but it's possible. We even don't need a compressor, many things a compressor does, can be done by EQs. Quality doesn't mean 2 Hz to 22 KHz and 200 dB S/N ratio. When using an analog tape 40 Hz to 14 KHz and 67 dB S/N ratio (EIAJ) and I take care about frequency issues and less good channel separation by the mix, I can do an amazing transparent mixes. _______________________________________________ Linux-audio-user mailing list Linux-audio-user(a)lists.linuxaudio.org http://lists.linuxaudio.org/listinfo/linux-audio-user

Thanks for posting this. I cannot tell the difference with ABX - even for the 108 version - so I guess either my ears are shot (quite likely) or my ears are not able to hear such subtleties (also quire likely)..

To my mind, a slight loss of quality (invisible to my ears) is a small price to pay in exchange for the much smaller file size...