OK, I'm a newbie to MP3 and trying to get a better handle on MP3 encoders. I am NOT interested in trying to make the smallest possible file size MP3s. What the heck than? I am just trying to make quality sounding MP3s with a reasonable file size, perhaps 1.0 MB to 1.4 MB /minute. On these forums I hear all this talk about not to use spectrum analysis, etc., just use your ears! Well I went ahead and did just that anyhow, using EAC with LAME 3.98 Beta 6 and iTunes 7.6.1. I have posted the results in a PDF file and provided the LAME and iTunes MP3 files for download at MediaFire:

MODERATION: Link Removed; please read the TOS before posting.

First I would like to say that I could hear absolutely NO difference between the ripped CD .wav file, LAME MP3, or iTunes MP3. From a file size perspective, here are the results:

.wav = 30.97 MB
iTunes MP3 (192kbps VBR Highest) = 4.278 MB
LAME 3.98 Beat 6 MP3 (-V 2 --vbr-new) = 3.71 MB

The LAME encode produced a 13% smaller file size than iTunes! That is not very significant (for me). Looking at the sonograms from Sound Forge 9.0, the original CD .wav has significant content up to the full 22 KHz. The iTunes encode retains most of content up to well over 20KHz, but the LAME encode starts cutting off at 16KHz, and everything above about 18.5KHz is gone. This is confirmed in the spectrum analysis on the last page of the PDF with the LAME curve (green) dropping off at about 16KHz, and very sharply above 18KHz. The iTunes curve (blue) follows the original (black) .wav curve, except for a few db drop above 16 KHz.

In listening tests the reverb tails and other areas where I would have expected degradation, the MP3s sounded identical to the original .wav. I can post the original ripped .wav file for comparison, but I'm primarily looking for an opinion on the two posted MP3 files.

I'm a newbie, so flame away!

If you can't hear a difference then you might as well go with the Lame mp3 file since it is smaller (15% is actually a significant decrease when you factor in 1000+ songs). Don't go off of the spectrum analysis alone as that doesn't matter, it only matter what your ears hear. You can do a spectrum analysis of iTunes AAC at 128kbps VBR and it would look terrible. However, most people would find the quality produced by 128kbps VBR iTunes AAC to be just fine. Your ears are the only things that matter here, not your eyes. Personally, I would trust a more developed mp3 encoder with true VBR settings over an aged mp3 encoder that aims more for bitrate rather that quality. Not only that but you know you are getting secure and accurate rips when using EAC, iTunes does not offer this.

JPEGs probably sound very different from the source bitmap images when played back as audio. My point? If comparing image formats by listening to them doesn't make sense, neither does looking at pictures of audio.

The idea behind perceptual lossy coding is that the audio is changed - dramatically - without affecting what you hear. If you graph it, a GOOD codec will look really, really weird but sound great. A bad codec will look just like the original, and probably sound awful. Make sense?

The iTunes MP3 codec generally speaking sounds like junk. Actually, I think Lame is the ONLY MP3 codec still under serious active development. If they sound the same to you, you COULD go with iTunes, but why not use the better codec? Don't worry about frequency rolloff... that's the codec doing it's job; removing the things you can't hear, in order to save space.

Your choice of lame -V 2 --vbr-new is a good one, and I would stick with it. (by the way, --vbr-new isn't needed with 3.98 and later, although it won't break anything). You might still want to stick with 3.97 though as its been tested more.

Looking at the audio is like listening to the image.

EDIT: Damn I was slow

There you are, then. As post #2 points out, if you can hear no difference between all the samples why wouldn't you go for the smallest one?


FWIW, listening tests that have been published here indicate that LAME MP3, when used at the setting you used, will stand comparison even with some more modern codecs:

http://www.hydrogenaudio.org/forums/index....showtopic=36465

That test was conducted a couple of years ago, but the general point stands.

OK, I'm getting the consensus that even at moderate compression levels and higher bit rates (i.e. 192kbs VBR High Quality) it is preferable to set a high-frequency cutoff at 16Khz to 18KHz, rather than extend the frequency response to the max. 22,050 Hz. This allows for more accurate MP3 encoding with less "audible" artifacts. Actually 16Khz is FM broadcast quality, which is not bad (from a freq. resp. perspective only!).

Any other advise on using EAC and/or LAME with iTunes as the synch program for my iPod Touch? I like the convenience of using iTunes all-in-one for adding music, videos and photos. BTW - the iPod Touch does a crappy job on photos everything is saved as a 640x480 bitmapped image w/o cmpression.

I guess I'll also try LAME 3.97 as well. I cut my teeth (literally) on 78 RPM records at age 3...way back in 1949! It seems funny now that vinyl LPs are now considered better quality than commercial 16bit 44.1k CDs. Thanks for the input...Like I said I'm an MP3 newbie, but new new to audio and technology.

If there is a difference in quality between vinyl and CD it is mainly due to poor mastering of CD. Otherwise it is easy to show that with equal care the CD can run rings around vinyl. (BTW, I too was born in 1946 and listened to my grandfather's 78's)

Please corroborate this claim with results from valid listening tests.

Right, you want to leave the Lame frequency settings as-is. You don't want to increase the frequency response to 22.05KHz as it would only introduce more audible artifacts. A -V 2 mp3 definitely has more sound quality than an FM broadcast. I think that an FM broadcast is similar to a 64kbps mp3 but definitely not anything higher. So leave the frequency settings alone, they were put in place for a reason.



Actually, photos (actually images) are stored on the iPod touch using Apple's format. It compresses the images quite a bit and reduces their resolution down to 640X480. A 640X480 bitmap image would definitely take up too much space so Apple implemented their own photobook technology. I too prefer iTunes for managing my audio but I use a program called dbpoweramp to do all my ripping. I make sure all the CD information is correct in dbpoweramp then I rip an audio CD once to Apple lossless. I then import that file into iTunes so that I have a lossless archive. I encode the file to 128kbps VBR AAC using iTunes and to -V 2 --vbr-new with Lame mp3. I have the 128kbps VBR AAC files for portable playback and the -V 2 mp3's to use with my PS3. I use iTunes to add album art. The iTunes store uses high resolution album art that is around 640X480 and it is either uncompressed or nearly uncompressed. I then embed the album art so that my other devices will use it.



Go with Lame 3.97 or the latest beta. You normally shouldn't run into any issues especially since Lame 3.98 is already up to beta 6. If I remember correctly, Lame 3.97 became safe to use all the way back when it was still in beta 1 phase. Technically speaking, CDs should have the same sound quality as vinyls. That being said, music producers are using different mastering methods today. They use a lot of dynamic compression so that they can "bump" the bass. This is the way the new MTV generation likes to listen to their music. They like to have their 2000 watt subwoofers and nothing else, they want to feel the bass when they listen to their music. That is why you can find some really good older CDs that sound just as good (or maybe even better) than their previous vinyl release. It all has to do with the way the CDs are mastered, it has nothing to do with the actual format itself.

Mono perhaps. But W-FM and MP3 can't be compared directly and IMHO you are not allowed to think according to Rule 8.

All this doesn't matter. The only thing which matters is what you hear. Lossy encoding is not about graphs and numbers...... even if it would cut off at 12khz, it wouldnt matter as long as it cannot be heard (it could be, but thats not the point).

The reason for why lossy encoders "save" at the ultrahigh frequencies? Well, because:

1. The higher the frequency, the more bits are required for accurate encoding.
2. The higher the frequency, the less audible it is (starting at 15-16khz, humans tend to not hear stuff at all with normal music). BTW: Most unexperiences people highly overestimate their hearing. 16khz to the listener is not "high" frequencies like hihats.... 16khz is way way above that. Normal music content ends at about 13-14khz... everything above that is just faint noise).
3. Resources are not infinit - you have a limited amount of bits, and then need to choose how to distribute them. 1bit spend in one area, is 1bit not spend somewhere else. In other words: adding bits to one area, is the same as taking bits away from another area.

I just want to add to this. There is a difference between hearing one single note at a certain frequency when compared to hearing that note with other, lower frequency instruments. I can successfully hear a single note at about 17-18KHz but I cannot hear normal music in that range. So there is a difference in being able to hear a certain frequency and being able to hear a certain frequency while listening to actual music.

OK, so it looks like LAME encoder is doing a very good job of setting cut-off frequency and applying filtering. I have posted the original PDF test results here to see this:

http://www.hydrogenaudio.org/forums/index....showtopic=62304

Is the consensus that LAME 3.98 Beta 6 is stable and there is no need to revert back to 3.97 final release version?

Lame vbr is tuned and 'stable' from v3.95 . IMO you can't go wrong at quality V4 or better with any of them.

No better then most any other encoder.



Before you waste time making any more of this junk, consider reading this thread. Its filled with people trying to talk some sense into you. Might want listen. You'll learn something that way.

Judging from you screenshot on the iTunes mp3 encoder settings you did not use Joint Stereo, you can really waste bits when using daul channel instead of Joint Stereo, while Joint Stereo IS very safe to use. A lack of a low-pass filter and Joint Stereo can really waste bits since you are very unlikely to tell the difference. Also I did find the VBR on iTunes Mp3 a few years ago to be really useless compared to LAME, but it could have improved since the last time I used it which was around 2005. But wow those settings on the iTunes Mp3 encoder really remind of that awful Blade MP3 encoder since that lacked a low-pass filter and Joint Stereo.

I appreciate all of the feedback provided.

I would like to respond to Mike Giacomelli's remark concerning "wasting anymore of my time" posting test results. I made the following statement, "Like I said I'm an MP3 newbie, but new new to audio and technology."

What I meant to say is I'm NOT new to audio and technology. As a design engineer who has been doing analog and digital circuit design since 1968, "testing" both HW and SW systems has been an essential part of my work for over 30 years now. I am also an avid musician who uses technology as a "means to a better end," not just to make test graphs to look at!

I started doing circuit design with vacuum tube technology in 1957 at age 11. So I have experience with technology that some have never seen, let alone know how it functions. Psychoacoustics is still a "black art," with bonafide certifications of vinyl LPs played back on Class A design zero-feedback vacuum tube amplifiers sounding better to many, than a carefully mastered CD copy created from original source played back on high-end semiconductor audio amplifier equipment. You can download the PDF of an article by Daniel H. Cheever titled “A New Methodology For Audio Frequency Power Amplifier Testing Based On Psychoacoustic Data That Better Correlates With Sound Quality” here on this topic:

http://www.next-tube.com/articles.php?arti...ub_menu_item=99

I will spare you all any further rambling, and hope I have made it clear from where I speak. Thank you again for your input and feedback. Cheers!

What people here are trying to tell you is that all that stuff you have been learning over the years is of little help when it comes to lossy encoding. Most of the rules are different and if you try to apply your old technologies to it then you will be led astray.

And please don't call psychoacoustics a black art. It is a well developed science and is at the very heart of lossy encoding. As long as you refuse to accept this you will never understand lossy encoding and what makes some encoders better than others.

Those who come here to learn are well served, but those who come here to teach generally don't fare so well.

Agreed, I am all ears (no pun intended)! I am totally open to any and all input here on lossy encoding. I accepted this from the getgo and said LAME is doing a good job, et al. Even though there has been no factual information presented, I am a believer in what you say through mine own ears.

But, if psychoacoustics is NOT a black art, what substantiation do you have of saying so. Also, what is "black art" and "scientific proof". Please look at the PDF article that I provided in the link above concerning this, and you may get a better understanding of what I mean. The article may be dated 2001, but the information presented is still valid!

Trust me, there is no no axe to grind - I'm just searching for a better means in creating MP3s or other compressed audio!

What do you mean ' no factual evidence' ??? Yeah and no black art is right as lossy has worked stable for years using mid~high bitrate. MPC has done it since 2000, mp3 since 2002 and now lower bitrates are competitive. The articles are here, just use search.

It's been a VERY long time since I last posted on this forum, but I feel that I may have something to add in this case.

I've spent the majority of my life as an electronics engineer and have designed and built a lot of very high quality audio equipment in that time. As you've proved for yourself Pixelpusher, what looks right according to electronic theory and what sounds "right" to the human ear are often two completely different things.

You know as well as I do that the feedbackless Class-A valve amplifier that you speak of would have been producing horrendous levels of distortion when compared to a carefully designed multi-stage amplifier with oodles of negative feedback yet it sounded "good". I've played with the same designs myself and am well aware of the "goodness" to which you refer.

If anything, you've already proved for yourself many years ago that it really doesn't matter to a large degree how much the output signal strays from the input signal as long as it either sounds the same as the original source to you or still sounds pleasant enough to be listenable to you.

It took me a long time to get used to the idea that an MP3 encoding could even vaguely resemble an original source signal when looking at the ways that MP3 treats the incoming signal. In the end I just learned to accept that if I couldn't actually hear the difference, which I can't personally at VBR -V3 or above, then any differences that existed were totally irrelevant to me. No amount of comparing graphs will alter that fact.

In the end, I just had to consider LAME as behaving in a vaguely similar way to a tape deck where perceptually transparent results are easily possible on high quality media despite much of the evidence pointing to the contrary when you try to analyse the system from a purely scientific viewpoint.

In the case of MP3, the only valid tests are listening tests where the MP3 encoding is blind-ABX tested against the original source material. When you find the settings where you personally can't hear the difference then, in your world, there isn't one.

Cheers, Slipstreem.

A fair enough request. They're on here someplace but I didn't bother to search... HA listening tests from a few years back.

There's Robertos Amorim's test where iTunes samples had a lower bitrate than all the other contenders:

There's also guruboolez's test from last year but was only done with Classical music.

Sebastian Mares put his ~128kbit VBR mp3 test on hold but recently expressed an interest in finally conducting it. The results should be interesting.

So people are interested after all in an MP3 test... Problem is that we never agreed on what settings to use for Helix / Real for example.

Discussion about details of Sebastian's listening test has been moved...

http://www.hydrogenaudio.org/forums/index....st&p=556303