The Great MP3 Bitrate Experiment

Lately I've been trying to rid my life of as many physical artifacts as possible. I'm with Merlin Mann on CDs:

This is a companion discussion topic for the original blog entry at:

I agree that 192kbps is fine the vast majority of the time but it still makes perfect sense to rip your collection to FLAC especially if you are dumping the original media. While we can sit and argue whether 192kbps or 320kbps is better, I’m sure most would agree that re-encoding a compressed file into another compressed format is a bad idea. That means if you have any intention of encoding those songs in a different format in the future it’s really in your best interest to keep a lossless copy around.

I agree about MP3 bit-rate too. But what if I throw a party and I have really big speakers? I heard that if you amplify an MP3 with really big speakers, you get distortion. Please some fancy DJ corroborate this.

the warez scene did a great job in figuring out the best mp3 encoding presets and this info can be found in their mp3 scene rules. You can check them out here Basically they choose LAME and VBR (best quality/compression ratio):

  • LAME 3.97 (final version!) with preset V2 and VBRNEW ("-V2 --vbr-new")
  • LAME 3.90.3 (modified version preferred) with preset APS ("–alt-preset standard").

Why is it so important to minimize the size of the tracks? Storage space is cheaper than ever. A 100$ 1TB hard drive lets you store 1,500 CDs in raw format. And the prices aren’t exactly increasing, either (at least not over the long term). So, you’re really not losing anything on ripping these files in raw format.

I would not spend time worrying over which bitrate to use, but rather just rip everything in raw format, and then if availability on portable media is an issue I would just make MP3 copies of these files to carry around on portable media.

I hate to be “that guy” but using the song “We Build This City” in the experiment is not really fair. That song was created with 80s drum machines and synthesizers. The quality of the original samples wasn’t very good to begin with. It would be like connecting an old NES to a stereo system and then complaining about it sounding just like mono.

Dave’s comment at #1 goes to the heart of the matter: For archival purposes, always use lossless compression. Re-encoding to a spiffy new future format MAY render some MP3 artifacts audible. And storage space is cheap. I have a collection of ~1200 CDs ripped to FLAC - it’s below 500GB, and it really doesn’t matter to me if this could be reduced to 100GB.

I am sure I can rate all of them in order consistently; I am also quite sure I could show people what to listen to in order to rate the versions correctly.

I will not do so in order not to impact the experiment, however listen carefully to:

  1. sibilants
  2. hats and crashes
  3. the small hi-hat that plays on the eights
  4. the splash cymbal (it only comes in at times)

I found these parts to change very noticeably in the 5 versions. There were only two versions which I had trouble placing: I was unsure about which was the third best or fourth best.

Listened on: iMac speakers and Sennheiser reference headphones. Didn’t need to bring out a proper audio card and reference headphones.

Adolfojp makes a very good point here. I really really REALLY like the idea of this test, but it’s flawed.

I grabbed an uncompressed WAV copy of the chosen track and gave it a listen for comparison, and found myself hearing what my ears identify as compression artifacts. I’ve trained myself to be able to hear them, which is probably a bad thing for consuming media, but it also means I get false positives on old tracks that use analog recording equipment and low-fi synths.

A much better track to use would be something like Aerials by System of a Down. It’s modern, mastered using high quality digital processes, has a huge harmonic range at both low and high amplitude. It’s my test track of choice.

Ignoring my own criticism of the experiment, here’s my reasoning for avoiding MP3:

  1. At low bitrate, it sounds significantly worse (to me, at least) than other codecs, such as AAC.

  2. Too many vendors of MP3 files use old versions of LAME, have it configured badly, or just use a low bitrate. 256Kbps MP3 produced from a poorly configured codec can sound worse than 160Kbps MP3 produced from a properly configured codec.

  3. I’ve noticed that MP3 has difficulty handling very complex audio, such as speed metal / black metal. For example, it tends to make that strange phasing noise over the top of heavy downtuned rhythm guitars during fast-paced high-frequency guitar solos. This may be my ears, or it may be the configuration of the codec, but I’ve heard it frequently (pun not intended).

  4. Every media device I own or have access to supports AAC, OGG or FLAC, all of which my ears seem to enjoy more.

  5. 1TB of storage space is ludicrously cheap. Heck, I’ve got just under 7.5TB worth of drives on my desktop machine. I could encode my entire collection in 32-bit 192KHz PCM and still fit it in there with plenty to spare. 320Kbps CBR AAC is more than small enough, and has a good quality margin.

Argh, typo: I user Sennheiser HD380 pro headphones, did not use AKG reference.


You mentioned that the playback fidelity of 24-bit/192kHz is slightly inferior to 16-bit/48kHz. What is the reason? After all you can map the 16/48 values exactly into 24/192, so it should be equal or better.

I think this audio quality experience poll is nice:

Well, except that it presupposes electrical music sources and (re)mastered originals, so it is rather hard (for me) to tell which artefacts were (deliberately) introduced in the original studio, and which ones were added/amplified by compression

I spent less than 15 seconds listening to the five different versions of the tracks and there were two clear winners (that I could listen to). I’m absolutely convinced I have picked the ones with the least compression (e.g. raw PCM and 320kps, allthough it is entirely possible the ‘second pick’ is in fact flac – I didn’t spend so much time vetting it)

Listened with puny Phillips in-ear buds and crappy work PC SoundMAX integrated audio… :slight_smile:

@Roman I can only venture to guess that audio equipment is actually better able to smooth out the aliasing and samplefrequency distortions at 44.1kHz than at 192kHz… I wouldn’t be surprised that driving the conus with sample ‘steps’ at 192kHz actually doesn’t work out that nicely (forcing the exact curve from hardware that just doesn’t allow for that?)

Anyways, you could follow the link to the research quoted

They all sound identical on my cell phone speaker on the subway.

I’ll agree with you that space is important, especially if you copy everything on a smartphone. That is why I keep everything as uncompressed audio. Huh? Well, mp3 is not the most effiency audio format. 256kbps mp3 is like a 128kbps AAC or ogg file. But what if you change your phone or music player? The iPhone don’t play ogg. So what are you gonna do? Rerip your ogg to AAC, and have double the compression artifact? Use mp3 and waste some space on your limited device?

Why not just keep the “lossless” on your computer and reencode it to your hearth content? For me, that’s THE reason to keep a lossless copy. That is also why I keep all my picture as RAW file, so that I can re export them all I want to any format I desire. Hard drive are so cheap these day… You don’t need to put that kind of file on a SSD.

@Roman it’s worth a read if a bit hard to get (even if you understand a lot of the basics). It mainly just points out that most hardware doesn’t have such a wide frequency range and is not perfect, so some of that (inaudible) higher frequency signal is going to bleed into the audible range and cause distortion, actually making the experience worse. I’m skipping all the details here.

Also 24 bit is unnecessary because psychoacoustics and proper encoders. So it just wastes space.

I’m sorry, but I have to say I couldn’t get past your second paragraph. You say this not being about piracy. Then, you mention donating the ripped physical media. Whether you buy, rip, then sell/donate or borrow, rip, then return the physical media, once you no longer have the physical media, you no longer have a license to the “software” and it would be wrong to hang on to the digital copy. I don’t know nor care about the legal versus non-legal copy argument, but this is just wrong.

Well, with only 5 samples of a single track there’s a chance of 1/120 of someone giving the right answer by random picking.

You’ll won’t be able to rule out that people with dog ears do have the ability to tell the difference between them…

I support the lossless strategy many have presented. I keep everything from Bandcamp in FLAC, and am considering reripping what few CDs I have (comparatively to most people) to FLAC. As for the rest of my collection, 80+% is in Ogg Vorbis, much of it from Jamendo. I often pass up downloading MP3 if another format is available, due to MP3’s age and technical inferiority. Storage and trnasfer is cheap nowadays. (until all your factories are belong to flood!)

@Romulo Did you consider that the poll was out of general interest?

Perhaps the poll shows that people clearly prefer a specific version - regardless of technical quality. Perhaps on average, people would think a 128kpbs is superior to a 256kpbs? Who knows.

Of course, that is not what we expect.

The test, however, is clearly not designed to identify individuals with perfect ears. It is therefore a poll, not a test.