Wednesday, March 14, 2012

'Mastered for iTunes is BS?' Bad Science Meets Rampant Retweeting

Mastering engineer and writer Ian Shepherd took exception to Apple's new Mastered for iTunes process being characterized in an interview as 'sounding closer to the CD.' So he took to YouTube and his website to prove that it does not. "Not true. Testably not true."

His result was re-blogged and re-tweeted without comment (2000+ times) until it not only became news, it threatens to become fact.

Unfortunately, his test design and logic are flawed, and the false conclusion he draws is the only conclusion possible under the test conditions. Let's take a look. [Disclaimer: I have no connection to Apple or its products.]

Basically, Mr. Sheppard used a well-known test that identifies differences between sound waves to compare a 16-44 WAV file ripped from a CD against two files.

  1. File A: Mastered for iTunes AAC of the same track professionally created from a 24-96 master, downloaded from iTunes store.
  2. File B: AAC created on his computer from the 16-44 WAV using iTunes with default program settings.
He then listenened to the files and concluded that File B sounded subjectively closer to the CD than File A. So it was "testably not true" (extrapolated from a sample set of 1) that Mastered for iTunes files sound closer to the CD than regular iTunes files. (Viewers of the YouTube video proof have to take his word for it; they can't hear exactly what he is hearing.) Test set size and possible equipment bias aside, I have two hypotheses as to why his conclusion isn't significant or surprising.

  • Hypothesis #1. Files A & B were produced from different sources. The basic iTunes WAV-to-AAC converter will create a unique result from any particular master recording of a song (varied by sampling rate, bit depth, equalization, etc.).
  • Hypothesis #2. An AAC file will have a stronger statistical correlation to its specific master file than to any other version.

If both hypotheses can be proven objectively, File B sounded closer to the CD because there was no other result possible. Mr. Shepherd's pronouncement is a tautology, a logical statement in which the conclusion is equivalent to the premise. And, as confirmed by the blogosphere and twittershpere,  another definition for tautology is "needless repetition of an idea, without imparting additional force or clearness."

My full test methodology can be found here. Using results from a sample set of three songs, we find the following:

Hypothesis #1. In all cases, the iTunes encoder produced a different AAC file from the 24-96 version of a song than from the 16-44 version, verified with the same null test used by Mr. Shepherd. The titles were chosen essentially at random from my 24-96 WAV collection and cross genres, so there is no reason to believe the result won't be repeated for any 24-96/16-44 file pair that does not have special characteristics.

Hypothesis #2. To get an objective measure of correlation, I mixed aligned versions of all WAV/AAC pairs (one in-phase, the other phase-inverted) and calculated the RMS volume of the resulting difference wave. That is, on average, how measurable the difference is between the WAV file and AAC file tested. Results below.

(Lower values indicate closer correlation between AAC and WAV files.)

In all cases, the closest correlation was between the 16-44 WAV and its corresponding AAC. This makes sense, as there is less information lost from compressing a 16-44 master to AAC than a 24-96. And in all cases, the off-diagonal files (an AAC mixed with a file other than its master) showed no closer correlation than the main-diagonal files (AAC mixed with its master), to the precision of the measuring tools.

Therefore, we can conclude that Mr. Shepherd's test was unwittingly biased toward hearing a closer match between the CD master and File B (created directly from the CD by standard iTunes) than File A (ceated by Mastered for iTunes from a different master). It is not surprising or significant that he did, in fact, prefer File B.

Nowhere in Apple's technical literature or advertising do they actually say the goal of Mastered for iTunes is to create audio files that sound 'closer to the CD.' Apple does say they set out to create a process and tools that produce audio files that are closer to the source. So we repeated our test method to compare Mastered for iTunes v. standard iTunes to see which produces files closer to a 24-96 master. Come back to see results.


            Vinyl-to-Digital Restoration #25            

Title: Ella Fitzgerald Sings the Cole Porter Songbook
Artist: Ella Fitzgerald
Genre: Jazz
Year: 1956



I didn't become aware of Ella Fitzgerald until the 70s, when her TV commercials. ("Is it live, or is it Memorex?") made her almost seem a caricature of herself. I'm very glad to have restored an original-edition copy of this LP from my father-in-law's collection to get insight into her earlier career. It reveals Ms. Fitzgerald delivering interpretations of Cole Porter songs that the composer himself must have thought true to his intent.

© 2012 Thomas G. Dennehy. All rights reserved.

2 comments:

  1. Do you think my test methodology is any better than Mr. Shepherd's?

    ReplyDelete
  2. Very interesting test.. Curious to HEAR/ read the results.

    ReplyDelete