Over on Macintouch, there’s a lot of discussion ongoing about music encoding — Which is better AAC or MP3? What bit rate is required for sufficient quality? How does the LAME encoder compare to the encoder used in iTunes 4?
Folks are expressing some fairly strong opinions. One popular post is someone’s personal comparison of music samples encoded with MP3 and AAC over a variety of bit rates. Sound quality is described using terms like “brighter”, “fuller”, “wider”, “clearer”, “deeper” and “more dynamic range”.
In my humble opinion, such reviews should be taken with a large grain of salt.
Some years ago, I worked as a student in the audio laboratory of Dr. Marshall Leach. During that time, I was amazed at the number of folks visiting the lab, claiming superior audio quality from things like gold capacitors, “high-end” cabling, and expensive speakers. What was even more amazing was that many of these folks continued to firmly hold their opinions in spite of laboratory demonstrations to the contrary.
A good example is loudspeakers. It’s actually not too difficult to build a near-optimal loudspeaker system — i.e. optimal in the sense that it produces a flat frequency response given a spectrally flat input (noise). Such speakers, which almost perfectly reproduce their input signals, are consistently rated poorly by audiophiles.
Anyway, back to encoding…
Today, I compared a music sample encoded on the Mac using the LAME MP3 encoder with VBR, high-quality, and specifying an average target bit rate of 128 kbps, with the same sample encoded in standard 128-kbps AAC using iTunes 4.
I listened to these two clips, running concurrently in QuickTime Player 6.2, and simply could not tell an audible difference. Physically the AAC file was 3.5 MB, compared to 4.6 MB for the MP3.
I would love to find a more technical comparison between these format. I would suppose that the such a comparison could be done through a frequency/spectral-power analysis of the data composing the song.
Has anyone seen anything like this available?
I read your entry on Macintouch, and I’d like to ask your ‘lab’ opinion on high-end audio cables. The best reason I can see for them is the sheilding to remove hum and other unmentionables from the signal. but is it true that their expensive wrapes and metals make better signal carrying?
Well, it’s representative of the music I listen to. That’s why I tested with it. (And with a few other electronic music tracks, getting similar results.)
As I believe I say on my web page, if you listen to a lot of orchestral music or indie guitar pop, the results you get might be entirely different, so you need to test the encoders with your music. For the music I listen to, AAC is terrible compared to MP3 (as encoded by LAME).
I wonder how much of it is AAC’s algorithms, and how much is the encoder, given how bad Apple’s MP3 is compared to LAME MP3. There’s an open source AAC encoder I’d like to try at some point, but it’s a pain to build so I haven’t gotten around to it.
Don’t believe AAC can be audibly worse than LAME-encoded MP3 at the same bitrate?
Check out http://www.xciv.org/~meta/audio-shootout/
I remember when studying digital signal processing, and specifically quantisation and encoding, that the design of such systems was all about trade-offs and context — what type of signal is to be encoded? Depending on the type (which therefore establishes a number of assumptions about the signal), we could perhaps encode the difference between samples, and not their absolute magnitude. Such encoding fails completely when the underlying assumptions of the signal source no longer hold.
Clearly the AAC sample you provide contains audible artifacts that are not present in the MP3 or OGG samples. However, the content of your music sample doesn’t come from natural sound sources, like voices or musical instruments. I wonder if your experiment is really “fair”, or representative, or whether it’s like saying, “If you really think JPEG is better than GIF, just have a look at this JPEG compressed computer-generated image of a white typeface set on a black background.” JPEG won’t handle those infinite frequency transitions too well, but then again, it wasn’t design to.