Currently most of my collection is in FLAC format, but the problem is nothing (other than PCs) plays that really, then the issue for portable devices is the size.

Anyway, I'm going to convert (but likely keep the FLACs anyway) the collection to either AAC or MP3 but I can't decide on which lossy codec to use. I've decided on either AAC or MP3 due to wide® spread support for those than any others, but I can't decide which would be best. I'm aiming for 10MB or less for most tracks but I'm not sure which of these is the best of the two when it comes to the higher end of the lossy spectrum, so I need advice.

I do have very good hearing and I'm a perfectionist much to my dismay which means I tend to notice artifacts and other anomalies rather easily and I'm going mad ABX testing different settings and encoders, so I thought I would ask you lot as you have considerably more experience with all this.

So, which encoder and settings should I be using?

Well . . . since you are the one that is going to be listening, I would advise that you do your own listening tests and determine which one works best for you personally.

Now... having said that, I believe most people would agree that the "safe recommendations" are -V2 for LAME MP3 and q.0.5 for Nero AAC. I personally use Nero AAC q0.45 for iPod use and I am satisfied with it.

Using Lame with -V 0 -q 0 actually produces files small enough, so would that be the best bet? I'm just trying to figure out which is likely to have least artifacts and stay more true to the original. When I try listening myself I become almost paranoid that I can hear things (that may or may not be there) which is why I'm also looking for reassurance about which would be the best choice.

ABX testing will give you a good handle on whether or not you are hearing things that are really there.

Yes, conduct a blind ABX test with software such as foobar2000. You cannot conduct a real listening test simply by switching songs back and forth or putting them in a randomized playlist. It is there that your mind could very well think it is hearing a difference. Make sure you use headphones to conduct the ABX test, use headphones that you would normally with your portable audio player of choice. Remember that portable listening environments (even with very good headphones) mask many frequencies and often don't let you fully hear the benefits of high bitrate encoding.

Lame mp3 and Nero AAC can give you excellent results at around the 190kbps VBR bitrate range. I believe the default recommended setting for Nero AAC is -q0.5 which will produce files around 170-190kbps VBR. I think the generally recommended setting for Lame is -V 2 which also produces 190kbps VBR files. Both should have an average file size of less than 8MB considering an average song length of 4 minutes (files will probably end up around 6-7MB depending on the complexity of the music).

Oh, and whatever else you do, DO keep the FLACs. It sounds like you may be re-encoding again in the future, and that's much easier if you don't have to re-rip, as well.

I wonder why there is always someone saying "V0 + q0"... ain't the HA recommendations enough?

I think AAC is much better.
You can see my post where I tried to compare both of them using a function (objective) metric. Here I have shown NeroAAC encoder generates files which are closer to the original file than Lame encoder does at any bit rate.
But not only this.
NeroAACenc can preserve the whole bandwidth of the spectrum - up to 22 kHz - at any bitrate above 200 kbps while Lame preserves no more than 20 kHz even @320 kbps CBR and around 19 kHz @220 kbps VBR.
And of course if you want better quality you need higher bit rate. Then, NeroAACenc can generate up to 400 kbps VBR (-q 1.0), while Lame generates no more than 320 kbps CBR and with the parameters of the best VBR quality:
-q0 -mj -V0 (--lowpass anything more than 20 results in 20 kHz bandwidth) - around 220 kbps.
Finaly, the VBR option should give the encoder the possibilities to encode the sound with the best quality for a given quality parameter wich is true for NeroAAC encoder but not for Lame.
Personally, I do not hesitate to use Nero AAC with VBR quality parameter -q 1.0 for me (~400 kbps) and -q 0.75 (~320) kbps for file sharing.

I was told that q0 became superfluous with LAME when using V0...

sure Sam, I read that too... but it's funny once in a while someone pops up and say "that they preferably choose q0 thing for a _much better_ quality." LOL

@SpasV: I don't think spectrum frequencies have nothing to do with quality? You want to see that? Play a song with loads of cymbals in foobar2000, turn on the equalizer and turn the 20 kHz up and down. Do you see any difference? If you don't, that's unlikely you will hear above 20kHz. That doesn't make LAME worse than AAC.

Pretty sure -no-one- can hear above 20Khz, I've never heard of anyone hearing over 18k, nevermind 20. My hearing caps out at 17350Hz and that's higher than pretty much everyone I know, if he can hear 19k then I'd be amazed.

The only reason I used -q0 was because it seemed like a case of "takes a little longer to encode but may improve quality, so as it's only done once you may as well not skip on it".

Where will you be using the lossy files? If it's for portable use, since you're keeping the FLACs anyway, I recommend using LAME -V2 and calling it a day.

First.

LAME: "Guide to command line options"
Quality related:

-m m/s/j/f/a mode selection
-q n Internal algorithm quality setting 0..9.
0 = slowest algorithms, but potentially highest quality
9 = faster algorithms, very poor quality


Second.

WikipediA:
"Frequencies capable of being heard by humans are ...
The range is typically between 15Hz and 20,000Hz"

"The human ear can nominally hear sounds in the range 20 Hz to 20,000 Hz (20 kHz).
This upper limit tends to decrease with age, most adults being unable to hear above 16 kHz"

It's Alive! Ultrasonic Spectra Isn't So Ultra Anymore:
(http://www.ocf.berkeley.edu/~ashon/audio/Ultrasonics.htm)

"When the consumer audio Compact Disc (CD) was introduced in the late 1980s,
the Redbook CD format specified 16 bit word lengths and 44 kHz sample rates.
This specification was sufficient for 96 dB of dynamic range and a frequency response
of up to 22 kHz (Nyquist principle). The limit of 22 kHz was based on consensus among audio engineers
that the human ear can only hear up to at most 20 kHz.
Limitations of the technology - of data density and in Digital-to-Analog conversion circuitry
- undoubtedly also influenced this decision."

And ..."Work done by James Boyk of Caltech ... has documented this over-20 kHz spectral content
in various musical instruments",
and "Additional work supporting James Boyk's findings was done by John Atkinson,
editor of Stereophile magazine, the preeminent audiophile journal.
In his October 2000 editorial he describes spectral analyses of audio recordings,
all of which demonstrate more or less activity above 20 kHz."

So, the frequency components up to 22 kHz are present in a CD sound no matter what I can hear.

this is why '-alt-preset--standard' and equivalent were so good. set it and forget it...

Yes but that type of logic doesn't get you anywhere. I can take it a step further and say that all lossy encoders are throwing away data that is present no matter what I hear. So that means that myself and everyone should use only lossless even if we can't hear the difference between a lossless file and a 128/160kbps VBR lossy file. That is the type of logic that gets you in trouble. Next you are going to tell me that you need a TV with a larger resolution than 1080p even though the human eye is limited to about that resolution. "Why there are pixels there that need to be in the picture even though I don't see them."

As for 400kbps+ AAC files, there isn't a single portable player that I know of that supports the playback of these files. So why have one 400kbps VBR and one 320kbps VBR library when the difference is only 80kbps (once you get up to bitrates that high, 80kbps doesn't add up much at all). Can you ABX the difference between a 320kbps file and 400kbps file? If you can then you represent the extreme majority of people.

I think the OP should just stick with -V 2 or Nero AAC at -q0.5

Yeah, I'm going to go with -V0 or -V2 with Lame. I was going to use AAC to start with but as I'm keeping the originals, I may as well go with MP3 due to its larger support (seems my XBox doesn't want to play AAC files, I thought it did but it says it supports "Playing AAC files from an iPod" so it seems to mean only that, not from Flash drives, stupid choice IMHO).

SpasV: Your post doesn't make much sense, I was arguing the point with the others about using -q0 as well as -V0/2. The second part is moot. Test your own hearing with some good speakers and a frequency generator.

The Xbox 360 plays AAC data files from an iPod or USB memory device. The thing it that it doesn't support Nero AAC, it only supports iTunes AAC. You need to optimize the Nero AAC files before trying to play them with the Xbox 360. The strange thing is that by default, Nero's AAC files are already optimized. That is why the whole thing is strange. At least this is a step better than the PS3 which doesn't support Apple's AAC tagging scheme and the Wii doesn't support AAC at all (or mp3 if you have the newest version of their photo viewing app).

I use both AAC from Itunes and Nero along with Lame MP3 and found that it seems that I can get the same sound quality with AAC in a 10-15% smaller file. That allows me to carry more music on my Rockboxed Nano. All do a very good job so I guess you might play around with listening to all and going with the smallest that sounds good to you.

If supported, I'd choose the AAC.

By the measures I've done, AAC @ 196kbps keeps the quality better than mp3 @ 320kbps. BTW, AAC gave excellent results even @ 128kbps.

Measuring results as Spectrum plots:

http://jiiteepee.fortunecity.com//tests/co...comparison.html

http://jiiteepee.fortunecity.com/pictures/comparison.html


I used Audition 3 for measures and the AAC for Audition for encoding/decoding.


Juha

Hmm, nice graphs, but where's the real evidence (ABX)?

You must have exceptionally good ears if you can tell LAME VBR -V2 (~190Kbps) from CBR at 320Kbps, Juha!

Even -V4 (~165Kbps) comes close to perceptual transparency for most people. Try reading the recommendations on the HA LAME WIKI rather than trying to expose faults through artificial means that are inaudible to the vast majority of people. That's how lossy encoders work. It doesn't matter a monkey's what a person sees on paper with their eyes. You listen to music with your ears.

Cheers, Slipstreem.

I'm sorry if those pictures "insulted" somehow ... my intend is just to show that there are differences between various codecs ... anything to do w/ what you maybe hear between those. You sure know the difference between hearing and seeing things ...


Juha



I think the evidence is shown on pictures ... you see there which codec keeps the quality of original sample best. Can you hear those differences ... that's not the point there.

Juha

I think that's precisely the point here personally, Juha.

You're trying to prove the effectiveness of a lossy encoder by throwing test tones at it that bear no relationship to the signals it's going to be asked to process in the real-world.

Now, suppose that I gave you a studio-quality cassette deck and a high-quality Chrome formula cassette tape to record your test tones onto. The manual says that the cassette deck has a -3dB point of 20kHz with this particular tape as quoted by the manufacturer.

How would you go about testing this? Would you record a series or sweep of test tones at 0dB and then accuse the manufacturer of lying when the figure comes out closer to 8kHz, or would you follow the industry standard for cassette decks of recording those test tones at -20dB and then come out with a figure as specced by the manufacturer?

Although that's probably not a particularly relevant analogy in technical terms, the point I'm trying to make is that there's a right way and a wrong way of testing things. Throwing hypothetical spot-tone test signals at a lossy compression system that's been optimised for handling complex waveforms in a manner that suits the human ear as opposed to the human eye is likely to give you results that don't reflect the encoders real-world sonic performance.

The only test that bears any real-world relevance where lossy encoders are concerned is what they actually sound like. That's why extensive use is made of ABX testing. Nothing else matters.

Cheers, Slipstreem.

By the way "AAC Plugin for Adobe Audition" uses FAAC 1.24.1 for encoding. So LAME should sound better. Anyway, lowpass value can tell you little about sound quality.

That is exactly the point here. You cannot visually represent something like audio quality. I have seen WAV plots that make 128kbps AAC look bad but yet audio quality is always about perception. You don't look at your music, you listen to it. The "evidence" in your pictures is nothing more than how good a graph looks. As I said, I have seen WAV plots that made 128kbps VBR AAC (encoded by both Nero and iTunes) look bad but the files were both perceptually good.

Please search the forums and you will see that visual graphs mean nothing when comparing the quality of lossy audio encoders unless you are actually plotting the ratings that each encoder/setting were given. Lame mp3 is also another example in that it too can produce some bad looking graphs even at -V 2 yet those samples turn out to be indistinguishable from the source lossless files.

So no plotting aloud as it doesn't prove anything. It is ABX or nothing.

QFT.