[ssamadhi97]

Originally Posted by robkismet

I've been meaning to make some kind of reference guide of MD (incl LP) and mp3 spectrums using white noise samples for months. Is this worth my hassle [..]?

White noise samples? That's not worth your time at all and has little use really. I mean, how many of us do spend their time on listening to white noise?

Seriously though, the specific fingerprints of lossy audio encoders are a direct result of the different ways their respective psychoacoustic models react to various properties of the input audio signal to exploit weaknesses in human hearing. White noise is only one very specific kind of input signal, which is why a reference guide based on it would be of little use indeed.

For example the changing lowpass you can observe on many lossy recordings is there because of the changing spectral energy distribution - and in some cases even the absolute amount of energy - in the input signal. Encoders can be distinguished to some extent by the lowpass they pick when encoding a signal of a certain distribution. When encoding white noise you only take one special case into account (namely uniform distribution)

Another example where encoders behave differently would be transients - some encoders (mp3, aac) switch between different block sizes to increase coding efficiency for tonal (long blocks) and transient (short blocks) material, while others ("old-fashioned" short-play md for example) use a fixed block size in all cases. Again, on white noise encoders will not exhibit any of this behaviour.


In short, you'd be better off using some snippets of real music instead.

Best you'd pick a bunch of clips that cover as much ground as possible wrt encoder behaviour on different signal types (ideally clips that cover tonal, transient and noise signals and different energy distributions all at once)