MP3 encoding quality: LAME vs MusicMatch - a test

[vin_petrol]

Chatting to a friend he didn't seem impressed that I still used WinAmp to listen to MP3 files. Apparently, MP3 files created by MusicMatch are of superior quality to those created by WinAmp. Actually, this was immaterial, as I use LAME to create MP3 files, and have done for years. I did try using MusicMatch a while ago, but it crashed with an error when I ran it, and the writers would only give me tech support if I paid the shareware fee for the more 'enhanced' version, which I was not willing to do. What with upgrading to XP recently it seemed reasonable to give MusicMatch another chance as an MP3 encoder.

The test that *really* mattered was as follows:
Rip and encode an arbitrary song from CD using two MP3 encoders:
1) LAME
2) MusicMatch
at 128kbps. Listen to the resulting MP3 files on my portable MP3 player and see which sounds best.

What with my hearing being less-than-brilliant after *mumble* years of loud music and motorbike riding, I didn't really expect to hear much difference.

The installation of MusicMatch went fine, although a BIG BLACK MARK to it for *automatically* assuming I wanted it as my default player for media files *without* asking me. This is something iTunes, WinAmp and even Windows Media Player did NOT do - all asked me if wanted them to do this ("no" being my consistent reply).

We also wanted a song that I didn't really know. Trash bought a CD recently by The Rasmus, which I hadn't listened to, so that seemed a good sample. We also wanted a song with "loud bits and quiet bits". She suggested a song off that called Still Standing, so I ripped and encoded it with both programs. I then copied them to my MP3 player and got Trash to randomly name the files (using a d6) whilst I was out of the room. I took the MP3 player away and listened to "2.mp3" and "3.mp3" for a while.

3.mp3 seemed to be slightly better. It had a kinda richer quality across all the frequencies, and was slightly nicer to listen to. Neither MP3 had any 'problems' per se - there were no clicks or buzzes or anything like that.

It turns out 3.mp3 was created by: MusicMatch. So my friend is vindicated, and I'll use MusicMatch to encode any future MP3s I create.

Comments

[blue-condition.livejournal.com]

It'd be interesting to try some others. I find "gogo" on Linux encodes very nicely, as does Mediaplayer 10. I used to use DAgrab on Windows, that just uses the LAME engine and is ok. iTunes does good VBR MP3s at higher rates but I think the MP3 codecs are just in there as a sop to communists who won't use DRM'ed stuff ;)

I've used WMP10 for most of my mammoth ripping sessions over the past few months.

[steer.livejournal.com]

gogo is simply batshit optimised lame I believe. I use it, it's pretty good.

[blue-condition.livejournal.com]

Yep, I know it has something to do with LAME but I'm not sure it doesn't have a different psychoacoustics engine in it too though. It's very hard to tell *what* is in gogo, given that it's VERY japanese ;)

[steer.livejournal.com]

From the lame site, "A forked version of LAME with many key routines rewritten in assembly (I.E, it's much faster). Uses the NASM assembler."

However, because of all that optimisation, it doesn't stay current with lame dev and hence is using somewhat dated acoustic stuff. Should be identical to an older version of lame.

[vin_petrol]

WMP10

Hmmmm. Good point. Mebbe I should try that too.

[steer.livejournal.com]

I can argue with this empiricism. You need a stat sig sample, not just two songs.

[vin_petrol]

Are you saying my methods are unscientific, Doctor Clegg! :-)

Over here, I have some cold fusion...

[steer.livejournal.com]

If you do a new test you should start from a .wav if you really do want to test the encoder rather than the entire rip/encode cycle.

[vin_petrol]

Really? What difference would that make?

This particular CD appeared as a set of 1 KB ".cda" files on my Windows machine, so I needed to use the inbuilt rippers of both bits of software to get at the WAV data.

[steer.livejournal.com]

I guess I just want to be clear what you are testing. Usually we can assume that reading from a CD is a straightforward "grab" of binary data but I know from ripping my CD collection that it definitely produces artifacts at that stage.

The ripping process is non-deterministic in the sense that if there are errors they may be different every time. Hence an error in your sound output might come from a CD misreading and rerunning your test with the same programs on the same track might produce different results.

However, since you would usually "rip" by inserting a CD and pressing go the test you probably want is to test the entire process since different CD error correction procedures might have different quality.

If you *really* want to try it I suggest the following test procedure: Rip and encode 20 tracks from different CDs using both (same tracks from both to reduce variability in the experiment). The tracks should be chosen at random from your CD collection rather than picked for dynamic range or musical "clarity" -- you will get some that you can't tell (I once tried to work out which rate to encode at using Portishead tracks which sound scratchy and fuzzy at the best of times). Alternate which ripper you use first in case there is a significant "cleaning" effect of being second ripped (e.g. if you are ripped second the CD is warmer or cleaner or something which affects read errors). Use a perl script to rename the original mp3 track names to A1.mp3 B1.mp3 with A and B assigned at random (but consistently so that A is always the same ripper) and the answer to which letter is which ripper is stored in a text file which you only look at AFTER.

[blue-condition.livejournal.com]

Just as an aside, has anyone managed to rip the latest Blue Nile album? No device or OS I've tried it on will rip it, but it'll play perfectly...

[nigelmouse.livejournal.com]

Have you tried a mac ? (if you've got access to one that is). I've seen impressive results with protected discs being put into a mac and then I tunes completely ingnoring everything apart from the audio, and popping up asking if the user wanted to rip it.

[blue-condition.livejournal.com]

Yes, I've tried ripping it on my Linux box (old LG CD-rom drive), my Windows machine (Samsung read/write anything combo el-cheapo drive), my office Windows PC (anonymous DVD reader/CD rewriter) and my Mac Mini...

It will *play* on all of the above.

No dice. :(

[cyberspice.livejournal.com]

I find that different encoders are better with different types of music. I think you need a bigger sample.

[vin_petrol]

I think you need a bigger sample.

One file is fun. A proper sample is hard work! :-)

Feel free to continue my work...