Lossy Audio Formats (mp3, ogg, m4a, etc), Quality and Comparisons
 

Index

• Introduction
• Explained : Analog vs. Digital Recordings
• Explained : Lossy Compression File Formats
• Explained : Constant and variable bitrates
• Question : Is one format better than others?
• Question : What bitrate should I get for the best listening experience?
• Question : Should I choose Variable Bitrate over Constant Bitrate?
• Summary

Introduction

Most of us by far have music collections on our computer these days and most of it are in lossy formats like mp3, m4a, ogg, mpc and more. Aside from potential differences between formats, the different formats also typically come in different bitrates representing the sound quality of the file. With all this to keep in mind, some questions may arise :

• Is one format better than the others?
• How high bitrate do I need to get if I wanna have the best listening experience?
• Is constant bitrate favourable over variable bitrate?

And so on. As I'm writing this intro, I don't actually know the answers to any of these questions yet, but I plan to look up sources and find out. I'll post my findings here and summarize them at the end. Before I do, let's just sort out some terms and expressions for those who don't know them.



Explained : Analog vs. Digital Recordings

An old-fashioned recording method like vinyl records sound as changes in a physical medium. Although it is an exaggeration, you can think of these formats as lossless. Conceptually, if you think of a needle scratching a vinyl surface, any sound waves will make that needle reverberate and those vibrations will be recorded in the vinyl medium. Although the representation is not exactly the same as the sound it records, it still allows for a very accurate representation of it. However, computers don't deal with physical mediums like vinyl, they deal with 1s and 0s. The "Yes or no" or "on and off" manner of which digital formats are recorded just doesn't allow for the same kind of representation that physical mediums do, therefore they are considered "lossy", meaning they lose some of the information you want to record.

http://services.exeter.ac.uk/cmit/mo...ue-digital.gif

Here you can see illustrated how the digital format is not able to accurately represent the soundwave that the analogue format represents. However, just like a digital picture format (like jpg) and how it's quality depends on the resolution of the picture, so you can raise the quality of a digitally recorded sound by raising it's resolution. When you look at the digital wave, everytime you see a 90 degrees bend or just a corner in the wave, that's a point where the sound has been recorded or rather - "sampled". If you drew a line between them, they would look like a wave. If you're familiar with maths, it's a bit like trying to calculate the area behind a line in a graph by the use of integration theory (squares basically), like in the graph below.

http://www.math.unh.edu/~jjp/radius/Riemann.gif

This is not a digital soundwave, however it illustrates the principle. If you want to accurately represent the area under the red wave, you can try and do so by filling the area with squares. In other words, the quality of representation depends on how many samples (squares) are recorded per time unit. This is known as the sample rate. The idea is that if you sample the sound often enough, you approximate the real "analog" sound. CD quality is 44100 samples per second (44.1KHz) and is regarded a good enough representation for practical use (although many argue still that vinyl is better, which they can be, although noise from dust and imperfect pins etc. may ruin the potential quality gain and who knows if the difference is even perceptible?).

There's also quantisation which basically means how detailed every sample is. In digital audio, quantisation involves representing the entire audible range of sound (20 Hz to 20 KHz) by a range of numbers. For example, audio CDs represent sound using a 16 bit signed number which denotes the numeric range -32768 to +32767. If we were to use a bigger range of numbers (which would require more data), then we would attain an even more precise representation of the original audio. (Thanks Seltzer!)



Explained : Lossy Compression File Formats

You can tell what the quality of your audio file is by looking at it's listed kbps, it's bitrate - or - how much information is processed when you play that file each second. A 128 kbps bitrate recording processes only half the data when played as that of a 256 kbps bitrate version of the same file. Also worth mentioning is that compressed file formats like MP3, OGG Vorbis, MPC and WMA are even more lossy than CD quality because they purposely discard all the information that we can't hear. In other words, a CD records "everything", also information you likely won't hear. Lossy file formats discard a lot of information (supposedly ~90%) to minimize file-size. How they figure out what to discard and how to do it (the "compression method") differs between file formats which means that there will be potentially audible differences between them.

Some lossy compression file formats : AAC, ADPCM, ATRAC, Dolby AC-3, MP2, MP3, Musepack, Ogg Vorbis (open source), WMA



Explained : Constant and variable bitrates

Lossy formats can either have constant bitrate, meaning they have the same quality from start to finish - or they can have variable bitrate which is an attempt to combine the best possible listening experience while preserving disk space. On a variable bitrate, the program that makes the file tries to figure out how much bitrate is needed to accurately represent the different parts of the songs it's recording. The higher frequency sounds there are and the more "complex" the sound is, the higher the bitrate needed to represent that sound becomes.



Okay, hopefully I was able to explain these things so we've got that sorted out. I'll now take a look at some sources and then post my findings below.

Question : Is one format better than others?

I've read several tests and will try to summarize the results I've come across. These tests typically pitch different formats against eachother in blind tests, testing the same recordings in the same bitrates, different bitrates and so on. It seems the differences are most noticeable at low bitrates and in general, a music library should not go lower than 128 kbps in quality for any format. Because 128 kbps is quite popular, I've summed here the results from a test testing OGG, WMA and MP3 at that particular bitrate.

1. OGG & WMA - 95 %
2. MP3 - 80 %

>> source : CD BURNER.CA - Comparison of Digital Audio Formats - MP3, WMA, OGG, WAV & more!

The percentages say how well they approach CD quality for the track they tested. As you can see, both WMA and OGG do better than MP3 at 128 kbps, but at higher bitrates, this can change and differences are often regarded as too miniscule to really notice. Note also that MP3 encoding may have improved since then, making it unsure if this description is accurate today.

>> source : OGG vs. MP3 vs. WMA vs. RA

There are noticeable differences at lower bitrates, but these low bitrates are mostly important considerations for streaming audio and not so much for listening to your local music collection. At lower than 128 kbps, MP3s tend to do bad and produces "compression artifacts", sounds that are not part of the recording. As such, other lossy compressions available at those bitrates tend to do better.



Question : What bitrate should I get for the best listening experience?

Of course, the higher you go - the better the representation. But when storing music on the hard drive, of course it's a compromise between quality and preserving disk space. The following graph shows mp3s at different bitrates and how accurately they represent the audio stored on the CD they are ripped from. We see herz along the X-axis and the closer the lines are to the red, the more alike they are.

http://www.fliptech.net/comp-all.gif

>> source : MP3 Producer Bitrate Comparisons - winamp bit rate mp3 mpeg3

As you can see, all bitrates fairly accurately represent the lower Khz audio-range and it's not until ~15Khz (which is within the audible range of most) that 112 bitrate file does a drop. At higher frequencies, it does not do a good job meaning information is lost. Because this is within audible range, 112 kbps (CBR) ripping should be frowned upon. 128 does quite a bit better and doesn't drop off until ~16Khz and I'm thinking that's probably right on the edge of what I can hear. The logical assumption here is that if you're like me and your hearing is not perfect, the difference between bitrates of 128 kbps and up compared to listening to the actual CD the files are ripped from is probably miniscule. However, as you can see on the chart - if your hearing (and equipment) is good enough (typical for you youngsters), you could potentially hear a difference between 192 and 256 kbps bitrate as there is quite a difference between those two bitrates at the ~17Khz .. provided you can still hear it. Also, what we see here and what I've gathered from other tests is that an increase of 64 to 128 will be a considerable change in quality whereas an increase from 192 and 256 likely won't because the difference then is only at the highest frequencies.

However, these are just charts and numbers. I wanted to see if I could find an actual listening test and see what that said. I've looked at tests where participants listen to files in uncompressed as well as compressed versions. These tests were all blind tests so the participants did not know the quality of the song they were listening to. The first blind-test had 4 people listening to 3 versions of 4 songs. One uncompressed, one mp3 encoded at 320 bitrate and one at 160. Participants are expected to guess right every now and then and the test seems to show that people (even audiophiles) are on average not able to distinguish between these bitrates.

>> source : Do Higher MP3 Bit Rates Pay Off? - Page 1 | Maximum PC

However, some other blind tests do report people claiming they were able to hear differences between 192 kbps and CD quality and 192 kbps and 256, but I've yet to find reports of someone hearing any kind of audible difference between 256 and 320. Looking at the charts, it makes sense that those with good hearing will be able to tell the difference between a 192 kbps and a 256 kbps bitrate.

>> source : MP3 vs. CD Audio Quality Tests & The Great MP3 Bitrate Test: My Ears Versus Yours - Gizmodo Australia

I did my own comparison just for fun where I listened to a 112 kbps mp3 to see if I could hear a difference from the actual sound on the CD. Remember now that I have no audiophile equipment and I have a tendency to say "huh?" when spoken to. I listened to Gartnerlosjen's song Kinosangen and slightly surprised, there was not much difference. The hats and cymbals that typically makes sounds in the high Khz ranges were crisper on the CD, but as far as the enjoyment of the listening experience went, it made basically no difference.

The answer to the question "what BR should I get?" becomes less straightforward, but here's an attempt to summarize :

• Recommended BR varies with the song. Classical music with fine string sections or electronica with high pitched sounds should be encoded at a higher BR. Rock, alternative rock, folk music and so on may not need the same bitrate.
• For young listeners with good hearing who want the best quality sound : 256 kbps
• For the average aged listener : 160 / 192 kbps - depending on the song. 192 should be a safe choice for most.

Question : Should I choose Variable Bitrate over Constant Bitrate?

Although it was harder to find reliable tests on this, it seems that in most cases, VBR is the ultimate compromise between disk space and quality. A VBR file will have better quality than a CBR file at the same filesize because the simple parts of the audio can be compressed better than in CBR, freeing up more bits for the complex parts. For the average listener, a VBR setting like 112 or 128 bitrate at the lowest up to 256 at the highest should produce files that closely approximate the experience of listening to the CD while preserving the most amount of diskspace.

Beware though that there is a margin of error and there is a slight chance VBR could produce the odd "artifact" in the sound or encode a complex part at lower bitrate than it should. However, most encoders have supposedly become so good at this that these problems are now unlikely. Hardware compatibility problems are now mostly a thing of the past.

Of course, the highest quality MP3s are of course CBR files at 256 / 320 kbps. However, the average listener is not likely to hear a difference between this and a 128-256 VBR file.

>> source : Constant Bitrate - Hydrogenaudio Knowledgebase



Summary

After reading through a number of tests, I've concluded with the following stuffs :

• At low bitrates up to 128 kbps, MP3 performs worse and may produce compression artifacts (sounds you don't want) more compared to other lossy formats like AAC, WMA and OGG.
• At higher than 128 kbps, differences between formats become miniscule
• VBR is a perfectly good way to preserve disk space while still enjoying high quality lossy files.
• Most people can't reliably discern 160 kbps bitrate from 320 and even uncompressed.
• 160 / 192 kbps (192 considered the "safest choice") should be good enough quality for most people.
• 256 or higher is only recommended for those with very good hearing and good audio equipment.
• Files with a VBR from 128 to 256 (though you could as well cap it at 320) should satisfy most quality-wise while being the most disk-space preserving option. It's likely a good safe bet for all.
• From a practical point of view, 256 and 320 kbps lossy files sufficiently approach CD quality enough to make "lossless" files like flac files an unnecessary waste of disk space.
• Also, lossy formats that can be encoded higher than 320 - no need.

Thanks again to Seltzer for sharing his wisdom and knowledge. That's it! Now it's time for comments, crass critique and - of course - any questions you might have. :)

Placeholder yos!

Very informative thread, I enjoyed reading it and looking at the charts and trying to piece two and two together. The science of sound is still something I have a hard time comprehending, it wasn't my niche for science (Earth science was more my thing), but I really enjoyed it. The chart of the bit rates was extremely accurate. Anything less than 128 kb/s and I discard it and look for something better.

Thanks for the feedback :D

After reading these tests and experimenting a bit myself, it's become increasingly appearant for me that I myself don't really need albums at 320 kbps. After playing too much drums without hearing protection and so on, I've become a sufferer of tinnitus and my hearing ability has taken a blow so .. Actually, 160 kbps is probably more than I need.

People with young ears like Anteater might wanna "safe it" with higher BR files though, if they wanna make sure they get the best. However, the listening experience will probably be as enjoyable with lower BR files.

192 kbps is the middle ground between the tired old ears and the audiophile's requirements.

Nobody "needs" it at 320 but if that's what's available then I'll take it. I'm not enough of an audiophile to really care, but FLAC format is just wasted space, I mean does an entire album need to be over 200 mb? for 11 songs?

lol, using calculus to explain digital recordings. but yes, it's a good analogy.

assuming i understood everything which was presented here, this is a brilliant fucking guide, especially for our audiophile nazis out there. personally, i can't even tell the difference as long as it's at least 128 kbps, no need for me to pursue lossless formats.

I can tell the difference between 128 and 320, but I can't tell the difference between 256 and 320.

Anyways top notch guide!

I always preffered v0 variable mp3 rips to 320, sound richer than the constant BR which can quite possibly be ripped poorly.

Yes. If I can get FLAC I'll get FLAC because it sounds better to the ear. 320 sounds horribly compressed imo.

Then you must have exceptional ears (you might have) because I've seen no test where people could tell the difference between these two reliably, only by lucky guesses. I'm guessing the "horribly compressed sound" of 320s is imagined.

You could blind test yourself, though. Put a CD with some songs you know well in your CDROM, then rip them to 320 and to 160. Try one song at a time and listen to them on shuffle so that you don't know which version is playing and see if you can guess which one is which.

If I'm not busy I'll get around to it tonight. It does seem interesting.

FLAC is 1411k! Are you sure you can't find a compromise? :D

Hah, discovered this old thing :p:

A friend of mine pointed out something that might be of relevance. He said higher frequency sounds that could be outside my audible range might create harmonics in lower frequencies. I'm not sure if this is something you'd want to preserve - you'd have to up the bitrate sufficiently to catch such interactions and maybe they're "noise" rather than something you want - but it's an interesting point.

great read, thanks for bumping this, i never would have found it :)

i always try to get music in the highest bitrate possible - preferably 320 or V0. i can't really tell a difference with anything over 256, but i figure with drive space as cheap as it is, i might as well. besides, some day i might have the cash to buy a real sound system, so i like to think of it as 'future-proofing' my collection while internet is still fast and music is still dreadfully easy to come by :D

I know that I can definitely hear the difference between different bit rates of the same piece of music. I think a lot of the time the difference is less perceptible if a greater degree of dynamic compression went into the original recordings, hip hop being a good example. As it stands right now I try to get everything at at least 256kbps.

I think it's easy for people to jump to conclusions. Reading studies on this, I have read instances where people believe they can hear a difference between bitrates and when they do a blind test, it turns out they can't. I myself turned out to be worse at this than I thought.

In order to be certain you can or cannot, you have to subject yourself to a blind test. Hearing a difference is not really the question either - the real question is can you hear which one is the higher quality one? I know that I'm not as good at this as I previously suspected and I think many would be surprised.

edit :

Actually, I can set up a blind test in the games forum, I'll encode a couple of mp3s in varying qualities and then people can download them and check if they can guess which ones are the correct ones. :)

.. Although they'd have to promise not to look it up on the file, though - or filesize. Hm.

I'd have to find a way to make people stream it I guess.

i tested myself once to see if i could tell the difference between a remaster and a regular recording. album was Pink Floyd's Dark Side of the Moon. i had 320 CBR that was ripped from who knows what, and a FLAC rip of the Mobile Fidelity Gold CD remaster. i picked the remaster correctly every time. i was really surprised at how clear the difference was.

but that's kind of :offtopic:

From my experience at the lower end of the bit rate scale you experience phase distortion which is that whirring white noise that most people associate with poor quality audio streams. As the bitrates progress further it becomes the sound stage and the dynamic range that is affected (mainly in the treble range, cymbals being a prime example), and the perception of distortion at dynamic peaks. I have done blind tests to be sure that there really was a different I was perceiving, but to me the difference between 192 and 256 is night and day. It should be noted that some people's ears are just more sensitive to this type of thing, being an audio engineer helps. Trust me, my hard drive wishes I didn't care.

I too hoard high bitrate files so for me the big difference is that I won't turn down a 160 if I can't find anything higher, hehem .. :p:

I have bad hearing, though, no way around it. In the studies, there are some that can pick the high quality files and these are people with good hearing who should arguably get their music in high BR. They seem to be a definite minority though. Satch, what you write is correct - one of the graphs I had in my post which is now sadly gone showed neatly how an mp3 up to 112 kbps could rather accurately represent sound of up to 15k hz or so before there was a rather steep dropoff point. 128k improved on this significantly, but would still create a dropoff within audible range for many if not most and the relationship between this dropoff and bitrate is not linear. An increase in BR does less for the quality the higher up you go meaning in theory there's a larger difference between 112 and 128 than there is between 256 and 272.

I definitely see this, but for me in most cases it's the signal to noise ratio and the soundstage, i.e. the stereo separation and the perception of instruments existing in 3-dimensional space that tends to improve beyond the point of 192. For me seems to be the pivotal point where the frequency range quality becomes imperceptible, yet other variables in audio quality are still apparent.

This is completely off topic, but depending on the quality and thickness of the vinyl and the type of music I can most certainly here a change between a record and a cd. I find it's hardly perceptible on music such as AC/DC, Iron Maiden, and other essentially straight forward music. But if I'm listening to something like Animal Collective's older work or School of Seven Bells, where there is a lot of different sounds in different frequencies that are all going on at once, it's quite easy to hear a difference. You might have pops and so on on older records, but where it's not scratched I think they'll still sound better. I haven't really tested with older records before, but I garauntee something like AC, SVIIB, or **** Buttons will sound better.

Back on topic of compression, I've done a blind test and can definitely hear a difference between 192 and 256. 256 and 320 I really don't hear too much of a difference at all. This could be because per Satchmo's comment, if the frequency range is more or less the same once you get to 256 kbps. But in all fairness, I can hear up to ~20 Khz, which I know because I recently had a hearing test for the Navy. There may be differences between 256 and 320 but I sure can't hear it.

regarding vinyl, i can tell immediately when listening to an mp3 if it has been ripped from vinyl. the sound is much softer, much more muted. this is especially true for electronic music. i prefer digital over analog any day.

Um... you do realize that an mp3 ripped from vinyl is digital right? Of course it's going to sound crappy if you rip analog to mp3, you are compressing the sound. Imagine a wave. An analog recording is just a straight, completely solid line, whereas a compressed file would be dots along the wave; the higher the quality the more "dots" or samples there are.

I really don't think there's any debate that vinyl recordings sound better than any mp3 compression...

if my surprise that a 320 CBR mp3 ripped from a vinyl source sounds astonishingly different than an equal bitrate mp3 ripped form a digital source makes me an idiot, then so be it. i remain surprised.

vinyl is fine for rock and jazz and whatever else, but electronic music started life in the digital world, and that's where it should stay. i never understood why DJs release instrumental stuff on vinyl, when CDs sound so much better.

It could be that more compression was added to the master before it was converted to digital for the CD pressing, but I have noticed that even homemade audio cassette recordings from vinyl retain that particular "fluid" quality that vinyl is famous for.

no, natural harmonics are always integer multiples of the frequencies which you already hear. so you might be missing out on some higher harmonics but they only go up from the base frequency.

the "noise" occurs when those harmonics are aliased -- meaning that the sampling frequency of the A/D converter (usually around 14.4 kHz, correct?) isn't sufficient to capture frequencies higher than ~7 kHz and aliases them to some lower frequency which manifests within the audible range. this might be what your friend was referring to.

Again, people who claim they can hear a difference - that's not really the question. So you can hear a difference between 192 and 256 - given 10 songs ripped in 192 and 256, could you tell which was which for the different songs with significance beyond the 50% chance you'll be correct for each attempt?

I mean that's what you have to ask yourself - not whether or not you can hear a difference. You could even play the same file to someone twice, tell them the BR is different when it's not and they'll come up with differences when there are none. Such is the human error.

When testing yourself, you have to know :

1. What are you after? You wanna be able to identify the higher quality file.
2. Your own weaknesses/error/bias as a human being. That's what blind tests are for, removing that error.

Sitting down listening to two files knowing the BR of both is likely not a reliable way to test this.

edit :

It's like my GF claimed she preferred Pepsi Max to Coke Zero .. I made her do a blind test. Know which one she preferred? Coka Cola Zero :p: Sometimes, people trust themselves too much.

I just wanted to say that the difference is identifiable, but when you are listening to a song on the laptop speakers or on the iPod earphones the sound quality won't be as good as listening to that same song on quality audiophile speakers. So if you're listening to a song of 256 kbps bit rate or an uncompressed Audio CD of 44 100 kbps on that same laptop speaker, you will hardly notice the difference.

Studies on several people (using mostly the blind test) have showed that a file compressed to less than 256 kbps is mostly noticed. It's not the case of every person, it depends on how sensible your ears are, and how much you trained your hearing.

Well I've done that before with a friend, trying to find out how high quality we really needed to rip cds. I could hear the difference between 192 and 256 every time, but after that point couldn't tell. So it's definitely identifiable, the thing is it's not a big enough difference that I actually care. I rip cds at 256, and download whatever I can find but preferrably 192 or better.

And noise, it definitely doesn't make you an idiot. I'm not arguing that a digital recording of a record will or should sound any better than an mp3 file. What I meant is that it's no longer an analog recording if you rip it to mp3. At that point, the music is digital. I was talking a comparison of an actual record played through speakers from a turntable versus an mp3 compressed file played through the same speakers on an mp3 player or computer or whatever. I definitely think the vinyl sounds a lot better.

in the end, i just say - why not? 90% of the time i can find what i'm looking for in 320 or V0, so that's what i aim for. $100 gets you 1TB of drive space these days, that will will hold an insane amount of music, even at these high bitrates. so i figure i might as well get HQ stuff.

there still remains the issue of lossless rips though. i usually only grab flac of things i want to archive. the new beatles remasters, for example. or the new kraftwerk remasters. i also spent an embarrassing amount of time tracking down flac rips of what are generally accepted to be the best remasters and/or best pressings of pink floyd's studio albums. i have lots of drive space, my internet is disgustingly fast, and music is very easy to come by these days, so again - why not?

A TB is a lot of drive space but unless you have a very high performance front end audio system and outboard D/A converter to boot, you could cut the bit rate of a FLAC file in half and it would still be well in excess of what anyone would be able to hear.

...which would suggest that audo cds offer much higher quality than is practically necessary. should we bring back cassettes?

i keep flac because it is a true copy of a CD, so if i ever want to burn CDs, i get accurate copies rather than compressed music. is that strange?

think digital cameras. today's point-and-shoots don't count for much, they use jpg compression to squeeze data down to reasonable sizes. but digital SLRs can shoot in RAW - the image equivalent of FLAC. my brother is an amateur photographer, and he always shoots in RAW. for most uses, he has to compress the images to JPG to use them, but he still keeps the RAW files around to have true and original versions of the images for archival purposes.

but yeah - you go your way, i go mine :D

Well, the music industry is selling mp3s nowadays .. what bitrates are they offering their customers? ;)

i have only bought a few albums online. most were flac, but one i remember was 320...

is that abnormal?

Yes, that is abnormal. The napster mp3 store which is one of the largest has most of it's catalogue in 256.

And Apples iStore which is an even bigger actor on the market ..

That was before, now they've stepped up to an equivalent of 256 kbps.

Source : Apple - Support

And having checked on bands who sell or give away their music, I think the latest Marillion album was available as 256 mp3s at the highest. Radiohead's In Rainbow was at 160 kbps while Maudlin of the Well's latest album was high quality VBRs.

So yeah, I'd say that as the industry is moving away from physical formats like CDs, they are abandoning excess quality. At least they have the last some years.

I took myself up on my word and made a bitrate test! Those who think they are good at identifying differing bitrates in lossy formats can give it a shot :)

Check it out -> http://www.musicbanter.com/games-lis...-bitrates.html

I'm definitely giving this a shot!

Me too, I do it for exactly the same reason. Have unexceptional sound card but I think there is going to be a difference when using old triple speaker system + a CD player.

Being meore into ethnic, jazz and avant, I think I can hear the difference. There's like more "warmth" to the sax part in lossless, for, example. Still, have to blind-test myself and probably will prove myself wrong.

A great article! Thanks a lot, tore !

For portability, I use LAME 3.98.4 at -V0 to encode my stuff - in the car or on my phone. I try to listen to 320 as much as possible.

At home, I can use use either MP3s (again, I try to use 320 but that's not always possible) or lossless...it doesn't bother me. When I'm downloading, I try to at LEAST get 320, but I can tolerate 256.

I haven't downloaded from iTunes yet but I would be willing to put up with their bitrates because it supposedly sounds as good as 320 does.