Are there any audible differences between 192 and 320 kbit/s .mp3 files?

Question

I have friends that are very picky when it comes to mp3 bit-rate, and will always look for the 320 kbit/s version of a file. However, I have never noticed any differences, they sound the same to me. I remember reading somewhere that the human ear is simply incapable of sensing the difference, even if present.

Are there any audible differences between 192 and 320 kbit/s .mp3 files?

Are there any good quality tests that have been done in this area?

What types of factors would need to come into play to tease out the differences (if any) in audio quality between a 192 vs 320 kbit/s .mp3?

---

This is a question that's currently being discussed on skeptics.stackexchange. It was thought reasonable to get some input from the audio experts. So with the author's permission I'm posting it here.

Answer 1

If you're after audible differences the only test you can use is a double blind test. Empirical tests such as waveform analysis will most definitely show differences in the output, but that doesn't mean the differences are audible.

In this case you could have some one listen to samples of music and have them guess the bitrate of the samples. The ABX type test seems to be popular for bitrate comparisons: you listen to sample A and then B and then X and decide which of A or B sample X is from. So A might be 192 kbps and B might be 320 kbps and the listener has to decide whether X is from the 192 or 320 kbps encoded track. This kind of test is a little better than an "assign the bitrate to a sample" type test because the listener can't guess their way through it with any accuracy and skew the results.

There are other double blind approaches, see this article for details.

There's been a few magazine attempts at uncovering codec and sample rate differences. This Maximum PC article gets a lot of attention on the net, though they're doing 160 vs 192 and 320 vs uncompressed.

I couldn't find any academic papers on the matter (but I didn't look that hard), which surprised me a bit. I would have thought there'd been a bunch of grad student research in psycho-acoustics available.

Answer 2

The ultimate answer for this is: It depends on what material you are encoding.

And the strongest scientific evidence is in the coders themselves. When encoding an mp3 using VBR (Variable Bit Rate), some encoders show how many frames were encoded using which bit rate. Here's a screenshot from LAME:

You will notice that only 10 frames out of 10735 were encoded using 320kbps, which is the highest bitrate mp3 supports. Had the quality been set to highest (which is not in this example) this would practically mean no audible differences should be heard. Had no frames been encoded using 320kbps, no audible differences should be discernible. Sadly, the output doesn't mention whether the 320kbps encoding was sufficient, or whether a higher bit rate would be needed.

In my experience, only dense and harmonically-rich masters put load on the 320kbps. Most vocal tracks, or bass recordings would rarely require more than 192kbps.

From various experiments that have been conducted, including some I have been part of, you should expect to find that our hearing is not aligned with the output of these coders. In other words, with everyday type of material and environments, people cannot tell the difference between 192kbps and 256kbps, let alone 320kbps; while many experiments are not comprehensive (testing environments and statistical analysis), there is strong indication that our perception is much worst than assumed by mp3 encoders. Bear in mind that most digital radio stations use 96kbps for transmission.

However, a final verdict should be made on a case-by-case basis using rigorous research, which is extremely hard to conduct. Alternatively, the process above should give you cheap and fast indication to whether it should even be possible for people to tell the difference.

Answer 3

I find it comes down entirely to the type of music and the speakers you listen to it through:

I can happily listen to pop music on an ipod through earphones at low bitrate (well, I say happily - not really a fan of pop music, but what I mean is it sounds as it should)

If you play the same mp3 through a decent or high end audio system it will sound worse, but again, with pop music it is still acceptable.

The real problem I have is in trying to listen to either heavy metal or classical music at low bitrate through a high end system. The jangles, splashes and other digital artifacting ruins the pleasure I get out of a good piece of music. This happens mostly at the top end (cymbals, distorted guitars etc) and especially when there is significant bottom end at the same time (1)

That said, I never bother with 320 constant bit rate because it just leads to large files. You should be better off with variable bit rate as it does intelligently reduce the file size where it isn't needed.

1- recording one of our new tracks found a major problem when generating mp3s - there are sections where not only are there a lot of high pitched distorted guitars, but some major subsonics that dive from around 80 to 15 Hz. That required top quality VBR to cope:-)

Answer 4

I work with sound for a living.

I have placed the same cut from 128, 192, 256 and 320 into a human ear sound analysis test range with a $100,000 sound sampling oscilloscope and the full wave form across the entire spectrum of the human ear is filled.

Further analysis shows that the main problem with mp3 sound samples in not in the bit but the distorted recording setting at any bit. Other words, people have a tendency to rip music at a height of input much higher then is needed causing envelop folding of all wave forms adding distortion in the applied recording.

Higher then 95% input is damaging to the recording as it leaves little overhead. My rule of thumb is to not record any audio over 89.2% as this will give your recording a level for overhead. All sounds have peeks and valleys which has to be given space for or its distorted by envelop folding for highs and crashes (base muffling) for lows.

To acquire a good recording you must stay within a channel of tones and then it is only replicated by the quality of the listening equipment and the receivers ear. Both of which by the way are never the same or ever perfect.

What I am saying is there is a heck of a lot more to the quality of sound then just the bit it is replicated at. Everyone is tone deaf at some level. You can play one sound clip to 100 different people and everyone of them will hear the same clip differently at some point. This has been proven in sound lab testes over and over. Then you put in the factor of human acceptance (like this sound or not) the best you can come up with from any recording is 88% acceptance.

So, you see, your fighting a loosing battle to try to explain the best quality. Just don't over peak your recording and let the individual receiver adjust the listening level for their own ear and equipment. Do your part by just sampling at a modest level. That's all that you or anyone else can or ever do to appease anyone.

Quality of input recording is the best answer as bit replication depth has little effect as long as it is kept within acceptable reason. Just my opinion after 30+ years experience.

Answer 5

There are some neat studies in the field of psychoacoustics that claim that lossy audio compression tasks our brains more than less lossy formats. I cannot for the love of me remember where I got this but I believe it was from TED.com I did do some background checking on those claims after the fact and they seemed to be fairly accurate.

Most of us can certainly tell the difference between 128 and 192 bit rates but the difference is less audible between 192 and 320. However, for the trained ear I purpose that there's a difference. I certainly feel that listening to some 320 bit rates compared to their 192 counter parts is less tasking. But it's really hard to apply this in the general sense. A lousy MP3 can definitely mess with my hearing even at higher bit rates but I cannot apply such a theory to the general population.

Listener experience (and training) definitely factors in. Unless your are running a streaming service and need to be conservative with the bandwidth, I say go with the higher, 320 bit rates.

Answer 6

To an untrained ear it would be borderline impossible to tell, and even a trained ear needs to pay attention. As such, it is hard to quantify perceptible differences as everyone will perceive it differently, if at all.

My approach to audio engineering, which I picked up from my time at SAE, is to go by feel. You can monitor everything, tweak to the nth degree, but as audio is inherently analogue, that is all you should really measure. As Ian C says, an ABX test is a great way to see for your self.

To answer your question from my experience the answer would be yes absolutely.

The degree to which you can perceive the differences will largely depend on the content of the song. Orchestral music is in my experience the perfect subject matter to test with. It is both incredibly complex in it's sound makeup (usually covering the entire frequency of human hearing) and has a very large dynamic range.

Bit depth also has an effect, my experience is that an larger bit depth 24bit as opposed to the cd standard of 16bit, also makes a great deal of difference. The next natural step up is vinyl.

As noted by some other people, the gear you use will also influence the perception. On my little white earbuds the difference is not nearly as instantly noticeable as on my large studio monitors. So if you were to test, you should use the highest quality gear possible to take harmonic distortion from the gear completely out of the equation. You could use a boom box from 1980 and not notice a difference between A or B.

Answer 7

Another thing to consider is who/how is/are they encoding it. I find a lot of the time that people are encoding and playing with settings they may not understand. This question is so finite though that it almost parallels the "analogue vs digital audio" debate.

Can it be perceived? Yes.

Will it be bothersome to listen to a properly encoded 192kb/sec mp3? Not likely...

Will it be bothersome to listen to an improperly encoded 320kb/sec mp3? Yes.

Is it personal preference? Mostly.

I can't listen to 128kb/sec mp3's... can't do it. Yuck! 192 is fine for me. Saying that, are there folks out there who have stereo systems that replicate the audio well enough that developing a preference for 320 over 192 is believable? Definitely... I don't have that stereo system though ={

Answer 8

Human hearing is great at pattern detection. MP3 has a psychoacoustical model trying to match the patterns in hearing as well as in the typical sources. A lot of transparency in music relies on phase information and MP3 is willing to bet on phase that is relevant and phase that isn't. A cymbal is very complex in its phase composition and to a good degree hearing is overwhelmed with it, so MP3 can simplify.

Another thing is that there is ambiance, reverbation and that drags down the phase coherence as well.

So the test objects that are most problematic are dry recordings without postprocessing and yet considerable complexity that takes a skilled ear to pick apart.

I have an accordion with a shallow three-reed tremolo of high quality. Accordion has very stable frequencies and high overtone content, and "tremolo" means that the (steel) reeds are slightly detuned with respect to each other to give a consistent beating effect. "Shallow" means that the beating is just few cents, and a 3-reed tremolo has one reed at a slightly lower frequency than the main reed, and one at a somewhat more high frequency than the main reed. A tremolo like that is tuned by ear since the precision of the beating and the compatibility of higher and lower beatings cannot be established with looking at isolated frequency readings.

So this sounds nice live. It's a bear to record in reasonable quality. And when you managed to create a good .wav, the threshold between intolerable and so-so for MP3 compression is somewhere between 160kbps and 192kbps. I'm pretty sure you could tell apart 192kbps and 320kbps. Apparently the psychoacoustical model of MP3 does not capture well what the ear catches onto in this case (Vorbis behaves more in line with expectations regarding the relation of audible quality and bit rate for this kind of source). And microphoning is already problematic as well. Use a registration without tremolo, and you can easily compress without impacting quality similarly obviously. Use an instrument that sounds awful to start with, and you'll not catch onto the difference easily. Even a 2-reed tremolo already works better.

So this is specific sound with specific elements where MP3 needs rather high bit rates to do a reasonable job. I'd imagine things like close-captioned jazz brushing techniques on a cymbal (complex noise with well-identifiable locatable components) would be similarly obvious in a side-by-side comparison but likely more in the "transparent/muddy" category rather than "nice/ugly".

As a general rule, "hearing music at 192kbps and 320kbps will make a significant difference" seems overblown to me. For most music, the difference will be if you try focusing on single instruments/components in a complex yet transparent whole and the original recording is good enough to make that endeavor successful. Which is more of a kind of sound engineer hearing than listening for enjoyment. I think it's rather rare for some sound to fall conspicuously apart under compression like that accordion tremolo.

Answer 9

See http://www.codinghorror.com/blog/2012/06/concluding-the-great-mp3-bitrate-experiment.html and http://www.maximumpc.com/article/do_higher_mp3_bit_rates_pay_off. Both are simple, but valid studies that indicate that we can't really tell any difference between files encoded above 160kbps.

A couple of caveats:

• The above figures are for stereo sound. For files with more channels (e.g. surround sound), you want correspondingly higher bit rates. So basically, you could hear the difference 160 and 320 kbps if there are more than 2 channels, as bitrate per channel will be lower.
• Bitrate is affected by bit-depth. A 16-bit sound file will be fine for listening, but if you're going to be mixing it with other audio, you might need the extra headroom afforded by 20 or 24 bit sound, and bitrate (and file size) will be correspondingly higher. But you won't be able to hear this difference by listening to the audio file on its own.

Answer 10

Plain and simple, YES, THERE IS A DIFFERENCE, PERIOD. No debate. If someone can't hear the difference it is because their hearing has been damaged and they just can't hear the freq differences. When you lessen the bitrate, you lessen the quality! I don't understand why this is even a question. LOOK at the data! Self explanitory. Beside that, there are a ton of results on google from MANY credible sources people! It's not rocket science....

Answer 11

I feel like a lot of people here are overthinking it and overexplaining in really boring details like they're kind of a "sound expert". I also wanted some answers about it. But going through all of the comments made it more confusing! Here is my experience on the matter.

The 320 format really is a big size for an mp3, so having your files with that will easily consume a lot of space in your hard drive. The formats I now use for mp3 to save space are 192, 160 and 128. If it goes below that, I can definitely hear a significant reduction in sound quality. But from 128 to 320, the sound is the same for me.

So if I acquire a file in 320, 256, etc., I use a program to turn it into 192, since I really don't notice any change or difference in what I hear. 128 is ok, but I have come to prefer 192 as my standard for mp3.

Now I have more than 20,000 mp3 files in my music library that I can enjoy listening to or explore at any time. I collect music, whatever genre or year, from as old as the 50's up to the present, from pop hits of every generation to the most obscure indie and music in other languages that are pleasant to my ears.

The bottomline is that from 320 to 128 the sound is fine or same. But below 128, the quality of sound is noticeably different or degraded.

Answer 12

I am a very picky listener. I notice that with acoustic instruments, you will hear the difference on a good audio set compared with mp3 players that are not high end audio.

I like dance organs from Decap Antwerpen (Belgium) and they are equipped with jazzflute ,vibraton, and flute. They have 8 registers that are natural pipes and the rest is electronics. It uses an inline synthesizer for sax, baritone, and trumpet.

I have all these music cd's stored as .wav and that sounds perfect. I tried .mp3 at different bitrates and is sounds less good with less ambiance... There is something missing in the warmth of the music.

Answer 13

Leaving the math out and measuring software off the table, there is some kind of distortion or oscillation in the middle upper frequencies. You hear it more with sustained Strings or Keyboard tone,and especially with symbols. All MP3's seem to have this issue no matter the resolution. It's less pronounced with a 320 kbps but still there. There is also a smaller stereo image in an open speaker situation. With headphones, I can't hear the change. A Wav/FLAC will be noticeably improved to an MP3 even at 320, with only a couple of exceptions being the type of music played, but even then, in an A/b test it will be noticed. How many people can, is another story. I was taught as an Engineer to remember only 5 percent of the population can hear what you hear, so don't spend all of your time making it perfect. Just worry about good.

Answer 14

Are there any audible differences? Yes, certainly but not a lot.

Keep in mind that MP3 cannot guarantee transparent compression at any bitrate. Not even at 320 kbit/s every song or sound effect will be perfectly transparent. There are some unfortunate sound combinations that just cause pretty much every MP3 encoder to fail even at 320 kbit/s.

The question is: Are there are lot of cases where the encoder will fail at 192 but not fail at 320 kbit/s? And therefor you have to take a look why an encoder can fail in the first place.

An encoder will usually not encode a signal it deems "unhearable" according to its psychoaccoustic model, no matter what the bitrate is, as that would only waste bits for nothing. The problem is: There is no perfect psychoaccoustic model in the world! All existing models have been created by making AB or ABX tests with voluntary testers and are thus only as good as the hearing capabilities of the test persons as well as the quality of the equipment used for testing. So if the model is flawed, it may throw out signals it should have kept. And even it is not flawed, it only models the average listening capabilities of the average listener. Yet there are people with listening skills far above average, too. If the model is causing the problem causing the audible difference, raising the bitrate may not help much or may not help at all, as the encoder may still throw away a signal it should have kept.

However, an encoder may also fail if it knows that a signal is hearable, yet cannot encode it as it ran out of bits. At some point the encoder has to also throw away hearable signals as the amount of signals it can store is limited because the bitrate is limited. Encoders usually try to be clever and throw away the signals that should be least noticeable to be missing but even if one of these are missing, some people may be able to notice that. And if that is the problem causing the audible difference, then of course raising the bitrate will improve the situation.

So the question is: How often does an encoder has to throw away important signals it couldn't encode at 192 and that it could have encoded just fine at 320 kbit/s?

To get an idea for the answer you can use a VBR encoder like LAME and use it as best quality (-V 0). At that level LAME tries its best to preserve every signal it considers hearable by human ears according to its model - it aims to be perfectly transparent to all listeners for all source materials. So whenever a frame is encoded with less than 320 kbit/s, then no more than the bitrate of that frame was required to preserve all these signals. Here's a sample output of running such a command:

Encoding as 44.1 kHz j-stereo MPEG-1 Layer III VBR(q=0)
    Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA 
  7800/7800  (100%)|    0:03/    0:03|    0:04/    0:04|   57.172x|    0:00 
 32 [   1] *
 40 [   0] 
 48 [   0] 
 56 [   0] 
 64 [   0] 
 80 [   0] 
 96 [   0] 
112 [   0] 
128 [   7] %
160 [  28] %
192 [  13] %
224 [ 829] %%%%%%%%%%%%%%%%%%%%%***
256 [4192] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%********************
320 [2730] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%***************************
-----------------------------------------------------------------------------------------------------------------------------------
   kbps        LR    MS  %     long switch short %
  274.4       77.1  22.9        85.2   8.0   6.8

As you can see, LAME was convinced that most of the time more than 192 kbit/s are required. If you had limited the MP3 to just 192 kbit/s, then most of the time hearable signals would have been dropped. Also notice that only in case of a 320 kbit/s block, LAME may have been running out of bits, as in all other cases, it could just have increased the bitrate if that was the case but apparently in some cases 256 or 224 were enough and in a few cases even 192 kbit/s were.

Note however that the example above was LAME trying to be as transparent as possible without just always using 320 kbit/s for file size reasons. Up to today, nobody has proven with an ABX test, that he can tell the difference between 320 kbit/s CBR and LAME -V 0 VBR MP3; which does not mean that people couldn't tell the difference between MP3 and uncompressed! It only means that when -V 0 was not transparent, CBR 320 kbit/s was not transparent either.

Yet most people neither have ears good enough, nor equipment good enough to require such a level of quality. For the vast majority of people and home equipment, -V 2 should already be perfectly transparent. So let me repeat that test again with this option:

Encoding as 44.1 kHz j-stereo MPEG-1 Layer III VBR(q=2)
    Frame          |  CPU time/estim | REAL time/estim | play/CPU |    ETA 
  7800/7800  (100%)|    0:03/    0:03|    0:03/    0:03|   63.878x|    0:00 
 32 [   1] *
 40 [   0] 
 48 [   0] 
 56 [   0] 
 64 [  27] %
 80 [  17] %
 96 [   4] %
112 [   5] %
128 [  14] %
160 [1554] %%%%%%%%%%%%%%***********************************
192 [3849] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%*************************************************************
224 [1167] %%%%%%%%%%%%%%%%*********************
256 [ 830] %%%%%%%%%*****************
320 [ 332] %%%%*******
-----------------------------------------------------------------------------------------------------------------------------------
   kbps        LR    MS  %     long switch short %
  201.7       41.4  58.6        85.2   8.0   6.8

And now you can see that to reach that level of quality, 192 kbit/s is in fact enough most of the time, often even 160 kbit/s is. There are only a few cases where LAME had do go up to 224 and only very little where it had to go all the way up to 320 kbit/s. So if I had limited encoding to 192 kbit/s, that would still have produced perfectly transparent frames to most listeners most of the time, just not to all listeners all of the time.

One of the differences between -V 0 and -V 2 is that the first one considers all frequencies as important whereas the later one only frequencies below 18671 Hz, which is one of the reason why it has to encode less signals. Yes, human beings can hear up to 20 kHz, but usually only small children. Already most teenagers cannot hear frequencies that high anymore and rarely any grown up adults, no matter how good their ears, no matter how good their equipment, so cutting these very high frequencies usually makes no audible difference for almost all listeners, unless these are young children and they will hardly care if these frequencies are missing in whatever source they are listening to. So yes, something is definitely missing here in the audio signal, regardless how many bits were used, but people who say they can tell first need to prove that by an ABX test and if such a test is made correctly, they usually cannot.

As for the psychoaccoustic model, one of the reason why AAC can achieve much better quality at the same bitrate or the same quality at a lower bitrate is because it uses a better and more fine grained model and encodes signals in a more fine grained manner, too, as one without the other one would make little sense. It would be possible to make a better model for MP3 but the format lacks the capability to make use of it. And it would be possible to design a model that works better even for trained ears but the result would only be encoders that produce bigger MP3 files in VBR mode with no benefit to the vast majority of users as they wouldn't hear any difference.

As a closing word, keep in mind that not only people with hearing capabilities above the average can sometimes not be fooled by the psychoaccoustic model, the same holds true for people below the average, as the model simply fails when your acuesthesia (= sense of hearing) is too far way from the one of the test persons used to create that model. And the later one is far more common case as the hard numbers prove:

Nearly 16% of adults in the U.S. report hearing trouble.

[...]

• About 40 million US adults aged 20-69 years have noise-induced hearing loss.
• More than 1 in 2 US adults with hearing damage from noise do not have noisy jobs, meaning the exposure is likely recreational.
• About 1 in 4 US adults who report excellent to good hearing already have hearing damage.

[...]

• 15% of school-age children (6-19) have some degree of hearing loss.
• An estimated 12.5% of children and adolescents aged 6–19 years have suffered permanent damage to their hearing from excessive exposure to noise, according to the CDC.

• Source: HealtHyhearing.com

So when entertainers who are regularly exposed to very loud stage music claim that MP3 (or any form of compressed music) sounds like crap and advocate for uncompressed digital music, take it with a grain of salt, as this is probably just because it sounds like crap to them. The fact that they can hear the compression artifacts while the majority of people cannot is no proof for excellent hearing. And if you cannot tell the difference, it doesn't mean your hearing is bad, it only means that your hearing matches those of the persons used to create the psychoaccoustic model, who were randomly chose but, of course, all tested in advance to not have any serious hearing defects.

So in the end it's also important to ask yourself one question: Who are you encoding the audio for? Just for yourself? Just for your family and friends? For a very large audience? Or for the vast majority of consumers? 192 kbit/s will be fine for most people and apparently also for yourself, otherwise you had not asked that question. If you encode for a large audience or for commercial purposes, you may prefer a higher bitrate, though, as this will even make some people happy whose acuesthesia is above or below the average.

Answer 15

I am new to this but thought it would be helpful to share my 2 cents worth. I noticed the depth of sound for a given music file when I stumbled upon an old version coded in 128kbps and listened to the same file back to back with a version coded in 320kbps. The annoyance is that once I sensed this difference, the perfectionist in me is now consigned to updating the remainder of my almost exclusively music library. Used reference quality headphones to notice the difference.