FLAC Forensics

Frequency Fingerprints

I was asked by a friend if it was possible to check whether some downloaded files were actually faithful to the original WAV format entities, or whether they had in fact been rehydrated from an unfortunate, lossy intermediate MP3 excursion. The files in question were FLAC compressions of the original 3½ hour, 42 track Analord series of electronic pieces by Richard D. James, most of which were only ever issued on 12" vinyl by the artist's now defunct Rephlex Records outlet.

Update: My personal RD James expert informs me that these original 42 tracks were also released in lossless digital format on Rephlex, and that some (or all?) of the subsequently released additional, digital-only tracks were also added to these by Rephlex.

I loaded the first of these files, SteppingFilter 101, into Audacity, and took a look at the frequency domain graph (Analyze|Plot Spectrum...). After a brief complaint about only being able to analyze 237.8 seconds of audio at a time, the result was this:

Notice the slight uptick at the extreme right (high frequency) end, around 22kHz. This represents extraneous noise generated by the digital sampling process, which in this case appears to have been set naturally enough to the CD standard stereo setting of 44.1kHz. This effect will be present in any rip, at some frequency or other, and is of a different kind from the artefacts introduced by MP3 processing.

Next, I used FooBar2000/LAME to convert this file to MP3 format, using the highest available quality, Constant Bit Rate standard preset, namely 320kbps CBR (actually LAME can handle non-ISO bit rates of up to 640kbps via its freeformat option, but very few MP3 players can handle such files).

The result of this 320kbps conversion has a very obvious steep cutoff at about 20kHz:

Any attempt to convert this back to the WAV format will preserve this telltale high frequency cutoff. Does this mean we can be confident that the source file represents a good, high quality, uncompressed rip from the original vinyl? I would say confident, yes; certain, well, that's another bottle of kippers. We haven't ruled out nonstandard MP3 or other shenanigans with this test alone, but those compression antics are at worst extremely unlikely.