Listen To All The Stuff MP3 Compression Leaves Out

[memyselfandus]

http://io9.com/listen-to-all-the-stuff- ... 1686279060

MP3s have become so ubiquitous that we often forget it's a compression format. When music gets trimmed to one-tenth of its original size, lots of information deemed "unimportant" gets tossed out. Here's what we're missing.

The video up top is called "moDernist." Ryan Maguire, a Ph.D. student in Composition and Computer Technologies at the University of Virginia Center for Computer Music, explains how he created it on his website, The Ghost in the MP3:

"moDernisT" was created by salvaging the sounds lost to mp3 compression from the song "Tom's Diner", famously used as one of the main controls in the listening tests to develop the MP3 encoding algorithm. Here we find the form of the song intact, but the details are just remnants of the original. Similarly, the video contains only material which was left behind during mp4 video compression.

Maguire says he used a 320 kbps bitrate because lower bit rates simply sounded more like the original, which would make sense given that more information about the music is being left out. Typically, a 4-minute song in uncompressed digital audio (usually a .WAV at 2-channels of LPCM audio at 16-bit sampled at 44,100 Hz) is about 40 MB. That translates to a rate of 1,411.2 kbps. After MP3 compression, a file can be squeezed down to about a tenth of the original, or a 4 MB size audio file; most mp3s are between 128 and 320 kpbs.

MP3s result in what's called "lossy compression"; data is lost and the quality is never as good as the original full bandwidth audio. MP3s reduce the accuracy of certain sounds considered to be beyond our auditory perception (which is why the method is often referred to as "perceptual coding"). For example, the human range for hearing is between 20 to 20,000 Hz, which is why CD quality audio is sampled as high as it is.

But as Maguire's project clearly demonstrates, there's lots of perceptual material that's getting tossed out.

[chk071]

Can't comment on the video, but i recently listened to the same track in wav and MP3 320 kbps on my brothers monitors, which are really good, and i wasn't able to sense the difference. He said he could hear it, i couldn't. Dunno.

[valhallasound]

You could do this with any track, a DAW and an MP3 convertor:

- Use the first stereo track of the DAW for the original high fidelity file
- Use the second DAW track for the MP3 compressed version, converted back to .wav/.aiff
- Make sure the tracks are lined up exactly in time
- Invert the phase of the second DAW track (i.e. multiply by -1)
- Listen to the results

[eXode]

Can't comment on the video, but i recently listened to the same track in wav and MP3 320 kbps on my brothers monitors, which are really good, and i wasn't able to sense the difference. He said he could hear it, i couldn't. Dunno.

It might be that he actually hears higher up in the spectrum than you. If I recall correctly, all MP3's regardless of quality has a "roof" @ about 16 kHz. Most people in their mid ages only hear up to about 16 kHz so they won't hear it, but some of us who are lucky (or however you want to put it) enough to hear up to 17-18 kHz will hear a difference, esp at the top end. It's subtle obviously but it's there.

[chk071]

Could well be, although he's older than me, and i should be the one being able to hear better. But well, i used to play drums, and always had one ear directed towards the snare drum, so i hear a little less on that ear already, so could well be that the higher frequncy spectrum is a little flawed already too...

[eXode]

Could well be, although he's older than me, and i should be the one being able to hear better. But well, i used to play drums, and always had one ear directed towards the snare drum, so i hear a little less on that ear already, so could well be that the higher frequncy spectrum is a little flawed already too...

Or it could be the monitors of course i.e. that they don't show the actual difference, and he's just imagining things!

[chk071]

He's surely had much more experience to listening with those speakers, i listened to them for the first time, maybe that's the difference. Wouldn't completely rule out the imagining stuff too though.

[J.Hartmann]

Could well be, although he's older than me, and i should be the one being able to hear better. But well, i used to play drums, and always had one ear directed towards the snare drum, so i hear a little less on that ear already, so could well be that the higher frequncy spectrum is a little flawed already too...

Or it could be the monitors of course i.e. that they don't show the actual difference, and he's just imagining things!

A blind A/B test never lies.

[Burillo]

just because you can hear a difference by reversing phase doesn't mean you'll hear it by doing ABX. you can delay a song by one sample, phase reverse it, sum with the original and "ZOMGTHEDIFFERENCE", but it'll be the exact same song, just delayed by one sample. that's of course not what happens with MP3, but it's a valid example of null test showing results that are meaningless.

[2ZrgE]

"MP3s result in what's called "lossy compression"; data is lost and the quality is never as good as the original full bandwidth audio."

Techically true, but other than that a purely academic discussion.

The important question is: Do we hear the 'degraded' quality or not (in an average listening environment)? With listening habits changing to the likes of Spotify/mobile phones/Youtube etc. I guess most people don't care anyways.

If people have the strong believe they can hear the difference between Wav and Mp3@320 - well, let them have their opinion, it's really useless to discuss with the 'golden ear' people (they never have to proof their 'abilities' anyways).

[BertKoor]

@eXode: that 16kHz roof is there only with bitrates 128kbps and below. With Lame at least...

Just wondering... the YiuTube audio with what was left out is encoded in mp3. So one wonders what was left out from THAT

[J.Hartmann]

"A recent study performed by audio researchers at DTS divided a group of listeners into two groups — one that watched a video accompanied by standard stereo 96-kbps sound (Spotify's default audio setting) and the other group listened in 256-kbps audio format. The responses in the brains of the group listening with the 256-kbps audio were 14% more powerful on metrics measuring memory creation and 66% higher on pleasure responses. And this was just 96 to 256 kbps. "

mic.com/articles/104250/what-the-internet-has-done-to-your-love-of-music

[kbaccki]

RIght, but 96 kbps is crap for anything useful. 320kbps is "very high quality" but still lossy. Who cares about 96 vs 256? The real debate is 320 vs. source. Saying "well, 96 to 256 is a HUGE jump in perceptive blah blah blah" is irrelevant -- what's the perceptive jump from 320 to source (if any)?

Also, Spotify is basically internet radio. How many people enjoy (or at least USED TO enjoy) listening to some FM stereo signal fading in and out, little fuzzy here, little scratchy there? Did it bother anybody that the difference between the source material and what's coming out at the other end of the transmission are different? Somebody agreeing to listen at 96kbps should just be told that it's low-quality, and that's that... sorta like a dicey FM signal. If you want higher bitrate on Spotify, spring for the premium account and stream at 320.