v1.2; last modified 2004-05-26 19:41:09 EDT by MikeSolomon

I've been ripping/encoding some more music recently and I thought I would re-explore the encoder field once again. Previously, I had encoded most of my collection with lame 3.92 using the --r3mix settings. I've used this under Windows (with ExactAudioCopy) and Linux (with CDParanoia) with reasonable success.

This has some slight bugs that are perceivable only infrequently, tends to "over-encode" a significant subset of my collection - namely punk. I perused the hydrogenaudio.com forums and despite the very low signal to noise ratio on that site, I've concluded that lame 3.93.1 is the build to beat. Unfortunately, even with some reasonable compiler tuning, lame can only muster about 2x realtime encoding with r3mix, barely 1x with their recommended setting alt-preset standard. This is all done on my 400MHz G4 TiBook.

I don't consider myself "golden-eared" but I am able to discern a lot of encoding errors that most people don't pick up, so the standard settings usually leave me cold.

I tried the following tests using iTunes 4.2 with all sound-altering code turned off (equalizer, sound-check, sound-enhancer) using only my Sennheiser HD-570 headphones in a quiet room.

Tested Encoders:

• AAC (QuickTime 6.4) 128kbps
• AAC (QuickTime 6.4) 192kbps
• MP3 (iTunes 4.2 128kbps Highest Quality Joint Stereo w/Smart) High Quality VBR
• MP3 (iTunes 4.2) 192kbps CBR
• MP3 (lame 3.93.1 --alt-preset standard) High Quality VBR
• MP3 (lame 3.93.1 --r3mix -B 256) High Quality VBR
• WAV (my control group)

Test Tracks:

1. Me First and the Gimme Gimmes - Where Do Broken Hearts Go
2. Sum 41 - Still Waiting
3. The Beatles - Sgt. Pepper's Lonely Hearts Club Band
4. The Toasters - Marlboro Man
5. The Toasters - Razor Cut
6. Apocolyptica - Path
7. Apocolyptica - Hope

Observations:

• in these tests, I judged 192kpbs CBR MP3 indistinguisable from WAV - thoughnotin a double-blind test environment
• iTunes VBR MP3 is terrible. Don't even bother. I cut it after two samples.
• iTunes AAC at 128kpbs displayed many of the same poor sound characteristics as poor VBR MP3 - warbling highs noticable in the cymbals and lack of clarity in guitar distortion overtones. I know "lack of clarity" in "guitar distortion" seems a little weak - but that's the best way to describe it.
• iTunes AAC is fine at 192kbps - it sounds as good as CBR MP3 at 192kpbs and has similar encoding times.
• given that MP3 is more widely accepted, I give the nod to 192kpbs CBR MP3
• lame VBR MP3 is indistinguishable from WAV in these samples, but comes with the following drawbacks:
1. encoding time
1. 192kbps CBR ~5x
2. lame --r3mix ~2x
3. lame --alt-preset standard ~1x
2. average file size - lame VBR is easily tripped up by the levels on a lot of punk/metal/rock cd's - it will encode up to ~240kbps average on most punk sample - conversely, you don't end up saving much on other calmer music that has been more artfully mastered - most samples averaged near 192kbps, with one or two were as low as 180kpbs

Conclusions:

I am an efficiency junkie which is why VBR encoding has a lot of appeal. However, after closer inspection, I think it's too easily tricked by questionable mastering techniques. Given the enormous CPU cost in encoding with lame, I think I'm going to switch to 192kbps CBR MP3. The aggregate convenience matched with excellent audio quality makes it my choice.

Future Tests:

I should throw in some techno - with some very sparkling, if not artificialy highs. Those wreak havoc with digital audio in general.

I will, of course, continue to monitor my collection and attempt to ferret out any sample that trips up the 192kbps encoding. Until then, I await iPod support for OGG Vorbis as the impotus for my next round of tests.