What is the best encoding format? Here's how to finally answer that question!
As you can tell from the length of this post, there is a lot involved in making such a determination, but I've never seen anyone try to quantify which format is "better" than another. So here's my attempt at it. Feel free to share comments, ask questions, or to tell me that I'm *--alt-preset insane* for trying to do this.
------------------------------------------------------------------------------------------------------------------------------------------------------
I've seen many posts from people asking or commenting on the "best encoding format". In my opinion, and in the opinion of many I think, there is no "best encoding format", only "favorite encoding formats" which vary from person to person.
You could use other factors to quantify a format, though, such as most ubiquitous (mp3) or best sounding (widely subjective for lossy's, but most would agree that any lossless format provides ideal sound quality).
A better question to ask is "What is the best trade-off between sound quality and file size?"
Ahhh...seemingly the question of the ages when it comes to digital audio encoding. Not many people would say "I've got a terabyte of storage space, to heck with file size, make it sound the very best!". (OK, stick with WAV...next question?) Nor would many people say "To heck with the sound, I'm legally deaf, give me the smallest possible file size." (OK, "lame.exe --alt-preset cbr 8"...nasty sounding, but each album only takes a few MB).
More commonly, people seek to maximize sound quality while making the most of hard disk space. The issue becomes: Exactly how important is sound quality to you in relation to file size? Do you have a lot of storage space? Not so much? Do you have "Golden Ears"? Or do you have trouble telling the difference between 1000Hz and 1200Hz? It becomes a question of EFFICIENCY.
Another important factor is compatibility. For example, I'd love to encode all my portable music into MPC, but my portable devices can't speak "MPC", so that's out.
Also, this calculation does not call for blind testing, as it addresses purely subjective considerations. It DOES, however, require a fairly stable test environment. In other words, don't listen to one format on one audio system and another format on a different system. Also, try to maintain an even volume level when listening to each encoding format of a given song. But don't use MP3Gain or WaveGain, as both of these methods change the nature of the sound by forcing some sounds below the noise floor. Best not to lose any sounds, in my opinion, but this is debatable for sure as is concerns sounds that would not be audible to most people anyway. Then again, I didn't design this equation for "most people", I designed it for all people, even those with the hearing of a dog.
Something else to consider is that many people don't use just one encoding format, but several. So you may be making several choices instead of just one when your testing and formulation is complete. Use the approach I describe in this post once for each platform you are testing for.
------------------------------------------------------------------------------------------------------------------------------------------------------
--- THE FORMULA FOR DETERMINING YOUR ENCODING SATISFACTION INDEX
ESI = ( ( ASQ ^ QW ) / ( AFS / AAD ) ) / 100
Legend:
ESI = Encoding Satisfaction Index ... To be compared to the ESI of each other encoding format
ASQ = Average Sound Quality ... Range: 1 (lowest) to 10 (highest)
QW = Quality Weight ... Range 0.2 (lowest) to 5 (highest)
AFS = Average File Size (in MB)
AAD = Average Audio Duration (in seconds)
Dividing the result of the fraction by 100 is to normalize the result into a value with a reasonable number of digits, but does not change the percentage of the difference between different ESI's. For example, 5800 and 4300 are an equal percentage apart from each other as 58 and 43 are.
------------------------------------------------------------------------------------------------------------------------------------------------------
--- PURPOSE
The purpose of this formula is to help you determine how much the combined benefits and caveats of a given encoding format measure in comparison with other formats. If you test only one format, you'll get a number which will be completely arbitrary without another result to compare it to. In other words, the purpose of this formula is to help you find the "best encoding format" for your own tastes.
The equation should be used once for each encoding format, with data you compile after listening to all of your test songs in that format. So if you're testing 4 encoding formats, plus WAV for reference, then you'll end up with five ESI values.
The results you get from testing and then using the ESI formula are for your own use only. The non-subjective factors are static (AFS and AAD), but don't expect anyone else to completely agree with you on any of the subjective factors (ASQ and QW). Since you're own subjective factors will differ from anyone else's, the ESI you end up with for each encoding format will differ from anyone elses outcomes. If high-bitrate MPC ends up being your favorite lossy format, that doesn't mean it's best for everyone. Someone else will hear things differently than you, and will have different priorities than you. For them, the silver bullet may be --alt-preset standard MP3, or Ogg Vorbis, And some people will only settle for lossless.
"To each his own." That's an important consideration for using this method.
------------------------------------------------------------------------------------------------------------------------------------------------------
--- DETERMINING QUALITY WEIGHT
To get accurate results, you first have to decide how important sound quality is for you in comparison to file size. AS I stated before, most people will not say "I only care about quality and not at all about file size", nor vice versa. Most seek the "perfect balance" between quality and size of their encoded music files. Assign a number between 0.2 and 5.0 to reflect how important you consider quality in comparison to file size. If you love high sound quality (and you have ears sensitive enough to tell the difference), BUT you have very limited storage space, you'll either have to give a somewhat lower number to QW, or decide that you can do with fewer songs encoded on your media storage device.
If you are making an ESI determination for more than one platform, you will probably have a different QW in each case. For example, my QW was 2 for determining the highest ESI, and therefore the best encoding format, for my portable devices. It'll be much higher for my primary (non-portable) listening devices.
A value of 1 for the QW would reflect an equal importance for you between sound quality and file size. This value might satisfy most people in the world who have an interest in sound quality, but also want to maximize the amount of encoded music they can fit on their storage media.
A value of 0.2 means, basically, "I don't care about sound quality, as long as I can at least recognize the song I'm listening to, just give me the smallest file size because I want to keep thousands of music files on the head of a pin!" Any number lower than 0.2 is not required to make the formula know that you really care MUCH more about file sizes than quality of sound.
A value of 1.2, for example, says "I care a little more about sound quality, and I either have plenty of storage space, or I can be satisfied with fewer encoded songs."
QW = 3 means "I care a LOT about sound quality, and I either have massive amounts of storage space, or I'll do with many fewer encoded songs but at higher sound quality than lower bitrates would provide.
Using a value of 5 is saying "I ONLY care about sound quality...either storage space is practically free for me, or I'll be OK with many fewer encoded songs on my storage media, AS LONG AS THEY SOUND THE VERY BEST! Now you're talking either straight WAV or a compressed lossless format. There is no need to use a number much higher than 5, no matter how much you value sound quality over file size. Remember that QW is used as an exponential in the formula, so using numbers larger by only one integer makes dramatic differences in the resulting ESI value.
For consistency, per platform use the same QW in the formula for all encoding formats, including the lossless baseline.
------------------------------------------------------------------------------------------------------------------------------------------------------
--- TESTING PROCEDURE
>>>1<<< Determine your preferred QW (Quality Weight) by referring to the previous section.
>>>2<<< Considering compatibility first, determine which codecs and which bitrates you want to compare. Now, you certainly can't compare all the formats and bitrates in existance, so select the most likely candidates in your own opinion...the ones you most want to form an opinion for, and of course the ones you have software to support. Also, consider whether you want to test the format(s) and bitrate(s) you are currently using against others, or if you only want to explore new possibilities. Most people I think would want to do the former in order to develop a point of reference. And certainly if you are asking the question "Is my format/bitrate good enough?" then it should be in your list of formats to test.
Make sure you select a good cross-section of codecs and bitrates which will work on all the digital audio players which you intend to use, even if you are considering multiple platforms each with it's own compatibility requirements. For example, I was considering using one format for archival and home/car listening, and another for my portable devices. This stemmed from the fact that my portables can only play MP3's, but I also wanted to consider higher sound quality codecs for the platforms that can play a wider range of formats (my PC and my car audio system). To keep this test managable, select between 2 and perhaps 5 codec/bitrate combinations. You need at least two in order to make the results of the formula meaningful (as one ESI by itself is meaningless), but for more than 5 it might take more time than you're willing to spend to encode all the songs you'll be testing. If you really want to test more than 5, though, obviously you can...the only limit is how long you're willing to sit in front of your PC and encode the same music files over and over.
And don't forget to include a lossless format for a sound quality baseline. This will make the results of the tests more meaningful, as you'll have one format with an ASQ of 10 to compare all the other codecs against. But, rarely will lossless provide the highest ESI, as the large average file size will tend to lower the ESI, unless you use a very high QW (Quality Weight), such as 4.5 or higher.
For myself, I only have compared several MP3 formats so far since that is the only format which my portable devices can play. Later, I'll test Ogg Vorbis, AAC, MPC, and some others to try to find a lossy codec which I can be happy with on my non-portable platforms, OR I may determine that lossless in the only way I can be truly happy...we'll see!
For my first test, encompassing the comparison of several MP3 bitrates against lossless, I compared WAV (as a baseline) against 192kbps CBR, medium-high-quality VBR, high-quality VBR and 320kbps CBR.
>>>3<<< Select ten songs which you think will provide good samples for codec/bitrate comparison. Pick some songs which are your favorites since these are songs you'll listen to more often, and therefore be more picky about the quality of. Also pick some songs which will be difficult to encode, i.e., full of harsh transients, wide dynamic range, many delicate passages contrasting with louder ones, etc. And include some songs which you consider "troublesome" for other reasons. For example, one of my favorite LeAnn Rimes songs, "The Safest Place", has a LOT of emphasis on her voice and not as much on the music. If you've ever heard LeAnn sing, you'll know how amazingly powerful her voice is...if you could turn sound energy into electricity, you could power a large city with her voice alone! My problem with this song is in having to turn down the volume to keep her voice from going right through my head, while still keeping it loud enough to hear all the music as well. Not so much an encoding issue, but simply a troublesome song in my opinion.
I'll share the list of songs I used and describe my criteria for selecting each one.
01 - Kalifornia (Fatboy Slim). This song is generally known to be a "problem child" for most any lossy encoder, expecially in the opening 20 seconds or so, with the vocal distortion wreaking havoc with most psycho-acoustic models. Great transients throughout, nice layering, fairly wide dynamic range and frequency range. A great test song for any codec and bitrate! Plus I'm a big Fatboy Slim (aka Norman Cook) fan, though I've only got five of his albums. I'd think this song is very difficult to encode well.
02 - You Were Meant For Me (Jewel). I've got a weakness for love songs, and this is one I like a lot. However, it's recorded at a very low volume (I think it was between 83dB and 85dB in an MP3Gain analysis). It's a very delicate song, but overall I believe it's not very hard to encode.
03 - Trip Like I do (The Crystal Method). What would an audio test be without some Crystal Method??? This song has a great variety of sounds, from the opening synthesizer, to a combinations of nice deeeeep bass hits and sustained bass boom, subdued vocals marking the first 2 1/2 minutes or so (the whole song is 7 minutes 34 seconds), decent transients, moderate-to-high dynamic range. Probably moderately difficult to encode really well.
04 - Love Me Like That (Michelle Branch/Sheryl Crow). This is a brand new song, on Michelle Branch's "Hotel Paper" album, and it has a great, fast acoustic guitar, emphasized strongly in the beginning. The voices of Michelle and Sheryl are quite strong, and the song has a decent frequency balance, with good bass notes and strong mids and highs. Encoding difficulty I would suspect is moderate for really good quality. Mainly, it's just a song I've listened to a lot recently.
05 - Right Now (SR-71). A LOUD song, recorded at just over 100dB according to MP3Gain, and very loud according to how far I have to turn down the volume to keep it in line with other songs. One of my favorites, this song has understated bass, over-emphasized electric guitars, and a very fast rhythm. I can turn this song almost wide open in my car without bass clipping...my ears will give up long before the system does. But overall, not a hard song to encode, I'm sure.
06 - The Safest Place (LeAnn Rimes). This is one of her newer songs (post-pop-crossover for her) on the album "Twisted Angel". She has the voice of an angel, but it's incredibly powerful and generally loud...I find myself turning down the volume to the point that the rest of the music's not loud enough. Probably poor engineering, in my opinion. I'm nowhere near as knowledgeable as a recording engineer, but I would have turned down her mic just a little and EQ'ed everything a little more evenly. This song mainly tests a codec's ability to reproduce a high voice frequency (2500Hz to 4000Hz) very cleanly, and to see how the well the codec keeps her voice from becoming harsh (I noticed lower bitrates had that effect, while higher bitrates made her voice a lot clearer). This song I would guess is moderately easy to encode.
07 - Time Out Of Mind (Steely Dan). Ahhh...my favorite Steely Dan song, track 5 from Gaucho. This is a great sound test song in my opinion, and I carry Gaucho with me anywhere I need CD audio to test a sound system (along with Fatboy Slim's "You've Come a Long Way, Baby", The Crystal Method's "Vegas" and Il Giardino Armonico's "Brandenburg Concertos" by Bach). The song opens with a powerful bass/snare/tophat that is, by itself, very easy to use for judging differences in sound quality between systems...or codecs/bitrates. Donald Fagan's voice is unique enough to justify using this song for an alternative vocal test, and the dynamic range throughout is respectable, as is the frequency range and number/intensity of transients. This song is fairly difficult to encode very well, not because it has very difficult passages (other than some persistant pre-echo), but because it's fairly easy to judge the quality of it between recording formats and platforms.
08 - Get Ur Freak On (Missy Elliot/Nelly Furtado). This song appears on the Lara Croft: Tomb Raider Soundtrack, which I would recommend to anyone who likes techno/industrial music, or just as a good sound test CD. Get Ur Freak On is not the best sound test song on the album, with BT's "Revolution" clearly being a better choice, but I wanted a song to really test sustained bass reproduction. Now, I generally don't prefer boomy bass, and very few of the songs I like has it, but this song is a good candidate for testing a codec's ability to reproduce rumbling bass sounds. This song has the kind of bass that people complain about as it booms down the street, and even my 500 watt car system with twin 8" subs can make my teeth chatter with it long before audible distortion or clipping occurs...I'm scared of what it could do on more capable platforms. But, for testing pusposes, I used it only at a moderate volume so I could hear all of the music without the bass drowning everything out. It satisfied my need for testing sustained low frequencies. Its transients, dynamic range and frequency range are average, if not a little lower than average (except for some decent bass transients). Not too hard to encode, I would think.
09 - Everything You Want (Vertical Horizon). Another of my favorites, this song opens with a great electric/acoustic guitar with good attack and mid-high frequency transients. Pre-echo is a problem with this song with the lower-quality encodings, and nothing compared of course to my WAV baselines. This song is also recorded loudly according to the MP3Gain analysis (I think it was at about 95dB or a little more, if I remember correctly). It's generally a moderate-to-difficult song to encode very well, mainly based on listening for its problem areas.
10 - Affetuoso from Concerto no. 5 in D Major, Johann Sebastian Bach (performed by Il Giardino Armonico). To satisfy my classical music sound test, I picked this one because I wanted something delicate. This song features a harpsichord, violin and a flute with changing overlaps, a great test for hearing each of the instruments clearly and with good balance and tone reproduction, though it's not a standard classical passage for having a very large number of instruments playing at once. I would think that encoding this song with high quality would not be very difficult, overall.
>>>4<<< Once you've selected your ten test songs, record all of them into lossless format for baseline comparison (like WAV, FLAC or APE), then into each of the other encoding formats you are wanting to compare. (Again, gain adjustments are not recommended for these tests.)
>>>5<<< Using your best quality sound system, listen to all the songs in each format and make each of the following determinations:
--a. Determine ASQ (Average Sound Quality). This step is CRITICAL. It forms the core of the decision you'll make about the format being tested. Your QW is already determined and is a fixed value for all formats and all tests. The average file size and average audio duration are non-subjective items which your opinion does not affect. The ASQ is the most important factor is the whole equation. This whole process doesn't make it any easier to form an opinion about sound quality, it only provides the means of quantifying your opinions into numerical values and using those values to obtain an overall satisfaction index (the ESI).
Assign a value of 10 to the ASQ of the lossless format you are using as a baseline.
Now, When you are comparing the other formats to lossless, don't accept anyone else's opinion, form your own. An ASQ value is specific to one person and to one encoding format. Develop an average opinion concerning the quality of sound for all songs in each format. Listen to each of the ten songs in the lossless format, then in the format you're assigning an ASQ to. Assign a collective number from 1 to 10 as the value of the ASQ (use decimals to be as accurate as possible). For example, if you think the overall sound quality of a certain format really stinks compared to lossless, you may assign a somewhere around a 2 to the ASQ. What you consider mediocre sound quality would be a 5 or so, and outstanding quality might me a 7 or 8 for a lossy format. However, giving ASQ a value of 10 for a lossy format means that it equals the sound quality of lossless in every way, at least to YOUR ears. This is very subjective, and the determination for ASQ you make with your own ears will differ from anyone elses determination for the same songs in the same encoding formats.
For the situation of two formats being so close that only some songs show any difference in quality at all, then reflect that with a very small difference in ASQ for that format. For example, if only three of the ten songs showed any difference in sound quality between two formats, and the difference was hardly detectable, then assign an ASQ of 0.1 higher (or maybe 0.05 higher) to the format that showed a slight improvement in three of the songs. This is a matter of your own opinion, but be as accurate and as honest as possible.
If some songs sound better in one format than another, but other songs sound not as good, then you'll have to determine what your overall opinion of the format is. Is the average sound quality a little better overall than the other format? Or a little worse?
And, of course, if there is no discernable difference to your ears in any of the songs between two formats, then they should have the exact same ASQ. Then it will be a matter of the smaller file size providing the best ESI between them when you're done.
--b. Determine AFS (Average File Size). The average file size in each format can be obtained by adding together the sizes of all your test songs in megabytes, down to two decimal places, then dividing by ten (the total number of test songs) and assigning the result (keeping two decimal places) to AFS in the formula.
--c. Determine AAD (Average Audio Duration). The average duration of the songs (which should not change across formats) is determined by adding the duration (in seconds) of all the songs recorded in a format, then dividing my ten (the number of songs you're using). Keep the resulting number in seconds and assign this as the value of AAD.
------------------------------------------------------------------------------------------------------------------------------------------------------
--- USING THE FORMULA AND INTERPRETING THE RESULTS
ESI = ( ( ASQ ^ QW ) / ( AFS / AAD ) ) / 100
Again, you'll be using this formula once for each encoding format, NOT once for each song. If you are testing five encoding formats (including the lossless baseline) for a platform, then you'll end up with five ESI values for that platform.
The ESI (Encoding Satisfaction Index) that you get when solving the formula means nothing by itself. It is only a reference point of how you generally feel about that particular encoding format based on your tests. The importance of the ESI comes from comparing it with the ESI from each other encoding format. It is meant to help you make an intelligent decision about which encoding format you should use on a compatible platform based on your own tastes and priorities. The importance of each ESI lies in the DIFFERENCES between them.
You must determine how big of a difference justifies changing formats. A difference of less than 5% would not, for most people, be enough of a reason to re-encode their music library. A difference of 50% or more probably would be. You'll have to determine how to respond to the differences in ESI based on your own interests. Is a difference of 10% enough to act on? 20%? It's entirely up to you.
Use the formula first with your test results of the lossless format, the baseline. First plug in the ASQ (Average Sound Quality) you determined for this format, then your pre-determined QW (Quality Weight). Now apply the AFS (Average File Size) and the AAD (Average Audio Duration). Now solve the equation! (Don't forget the proper us of the parentheses...just entering the numbers in order on a simple calculator will NOT give the same result as using the parentheses in the formula.)
The outcome will be your own Encoding Satisfation Index for that particular encoding format.
Now, repeat the use of the formula for each format you tested, using the appropriate values for each factor (the only one changing between formats should be the ASQ), until you have an ESI for each format.
Now, looking at the ESI's side-by-side, you can begin to determine how much of a difference is a "big enough" difference between the formats. If you determine that an ESI for a format you are not using is sufficiently higher than the format you are using, then that probably means you should re-encode. Or, if one of the formats has a significantly higher number than any other, then that is the best format for you out of all formats you tested. Keep in mind that to determine the best format for you out of all formats and bitrates in the world, you'd have to test all formats and bitrates in the world. This obviously isn't feasible. That's why selecting the most likely candidates is so important.
Use percentages to compare different ESI's.
ESI #1 = 15.92
ESI #2 = 13.01
Difference = ( ( ESI#1 / ESI#2 ) - 1 ) * 100 = 22.367% (ESI #1 is 22.367% higher than ESI #2)
Remember that if you reverse the numbers, you get a different percentage as you are now using the percentage of the other number. It's the nature of percentages...it's all in the order you say it.
Difference = ( ( ESI#2 / ESI#1 ) - 1 ) * 100 = -18.279% (ESI #2 is 18.279% less than ESI #1)
Just to show the extremes of what the formula will produce, let's try a couple of unrealistic theoretical situations.
Let's say there a format which, at a tested bitrate, sounds really horrible to you overall, in fact, it's pretty much static. But it doesn't matter because your quality weight (QW) is a 0.1. We'll use a *very low* ASQ of 0.1 to reflect your dissatisfaction with the sound quality. The average file size (AFS) is huge for this format, say 50 MB for a 2 minute song, which happens to also be the average audio duration (AAD) in your test. OK, let's determine the ESI...
ESI = ( ( 0.1 ^ 0.1 ) / ( 50.0 / 120 ) / 100 = 0.019
That's lower than any real world ESI you'll ever see, mainly because if there were a 25MB per minute audio format out there, it would sound a lot better than 0.1 ASQ!
Here's the other end of the spectrum. Let's say sound quality is VERY important to you (QW = 5). A theoretical format which you think sounds perfectly transparent with the source recording on every song (ASQ = 10), only needs 1MB to store 5 minutes of music. AAD = 300 seconds. Here's the verdict...
ESI = ( ( 10 ^ 5 ) / ( 1.0 / 300 ) / 100 = 300000
So, based on the ESI's, you could say that you prefer the latter format by 1578947268% over the other format.
In reality, for most encoding formats, you'll be dealing with differences in ESI of as little as 5% or so, and rarely as high as 200%. But, it's an open-ended formula, and the limitations I recommend are not stricly required. So if you want to say that your Quality Weight, for example, is 10, then go for it. Just be prepared for some dramatic results when you do the final calculation. If you insist on having a QW of 10, then you'll be recording your entire collection in FLAC or APE or LA or one of the other compressed lossless codecs, because if quality is THAT much more important to you than file size, then one of those formats will have the highest ESI for you anyway.
The reason I recommend factor ranges for QW and ASQ is to try to normalize the resulting ESI a little, while still making it expressive of your overall opinion of an encoding format.
------------------------------------------------------------------------------------------------------------------------------------------------------
--- THE RESULTS OF MY OWN TESTS
For comparison, here are the five ESI's I came up with based on my search for the best MP3 format for my portable devices. I used a Quality Weight (QW) of 2 for my own preference.
As for ASQ factors, I determined the overall quality of 192 kbps, my bitrate of choice prior to this exercise, as simply not measuring up to the other bitrates, and definitely not comparable to lossless. It had some audible problems with the intro to Kalifornia, slightly weak/muddy transients on Time Out Of Mind and Trip Like I Do, somewhat artificial sounding vocals on The Safest Place, a little weak and artificial sounding guitar attacks on Right Now and Love Me Like That, and not quite enough overall clarity and definition on Bach's Concerto no. 5 Affetuoso compared to the other formats. As a result of comparing it to the WAV's, I gave it an ASQ of 6.
For Dibrom's LAME settings, I found the quality of --alt-preset standard to provide a significant step up from the 192 kbps format I encoded with Windows Media Player. I detected differences in all of the test songs, and the overall sound quality was up significantly. After comparing it to the WAV baselines, I gave it an ASQ of 7.
The extreme and insane settings provided very slight improvements in sound quality over standard, with some songs being indistinguishable between the bitrates. But, that's why you want to average the ASQ, because you want to take the average into account to accommodate such variances. If the overall difference between formats hardly exists, then the ASQ you determine should reflect this. I gave the --alt-preset extreme format an ASQ of 7.1, and the insane format a rating of 7.2.
Here are my final results...
WAV:
( ( 10.0 ^ 2.0 ) / ( 41.07 / 244 ) ) / 100 = 5.94
192kbps (CBR) MP3:
( ( 6.0 ^ 2.0 ) / ( 5.734 / 244 ) ) / 100 = 15.32
158% better than WAV. Now this doesn't mean 192kbps MP3 sounds better to me than WAV, it means I prefer 192kbps MP3 over WAV for my portable devices, WHEN ALL FACTORS ARE CONSIDERED: average file size per average song length, how important sound quality is to me (QW), and my opinion of the average sound quality (ASQ) of the two formats. It's a judgement of efficiency.
--alt-preset standard MP3:
( ( 7.0 ^ 2.0 ) / ( 6.978 / 244 ) ) / 100 = 17.13
An ESI 12% more than the 192kbps ESI, and almost 188% more than WAV. A considerable improvement in efficiency overall.
--alt-preset extreme MP3:
( ( 7.1 ^ 2.0 ) / ( 7.727 / 244 ) ) / 100 = 15.92
An ESI about 7% less than the --alt-preset standard format, because of the marginal quality improvement and the larger average file size. If quality was even more important to me than a QW of 2, then I might prefer the extreme encode setting. This ESI is 4% more than that of 192kbps, and 168% more than WAV.
--alt-preset insane MP3:
( ( 7.2 ^ 2.0 ) / ( 9.231 / 244 ) ) / 100 = 13.70
An ESI 131% higher than WAV, but 11% less than 192kbps, 16% less than extreme and 25% less than standard. Sounds slightly better that the others in general, but inefficient in my opinion compared to standard.
If I had selected a dramatically higher QW, such as 4.0 or 5.0, then I'd end up with much higher ESI's, in the thousands. But keep in mind that you're not looking at one ESI by itself, but rather you're comparing the percentage of difference between two or more ESI's to determine the best encoding option for the given situation, all using the same QW for one test cycle.
Now, where are those Memory Sticks...I've got some tracks in need of re-encoding.
------------------------------------------------------------------------------------------------------------------------------------------------------
Next, I'll post the results of the tests for my primary platforms. For those tests I raised my QW significantly (as sound quality is even more important to me since storage space is at less of a premium).