Harmony Central Forums
No announcement yet.

Are your Statistics Any Good? or ...


  • Time
  • Show
Clear All
new posts

  • Are your Statistics Any Good? or ...

    ...The Delta Blues.

    I have prepared 4 statistical reports for the same longish 96/24 .wav file. SF and Audition reports look alike in key areas. Can any of the scientists, engineers and psychologists who hang out here relate the last two reports to the first two? thanks!

    Sound Forge 8 says:

  • #2
    OK, you have 3 different scales for amplitude going on here.
    For Audition the sound samples will have amplitude values ranging from -32768 to +32767 - - even though it reports it as a 24 bit file, this shows up as 16 bit values with fractional increments (-26233.11) - - if it were 16 bit, there would be no fractions.

    If you take any sample value and drop the sign so everything is positive numbers, you can convert a sample value to (instantaneous) dB:
    dB for a sample value = 20 log(sample value/32768)
    Using the Audition largest peak value:
    -1.31 dB = 20 x log (28173.58 / 32768)

    to convert the dB back to a sample value:
    sample value = 10^(dB/20) * 32768
    ( ^ means 'raised to the power of' )
    you lose the sign of the sample value (both negative & positive peaks can hit 0 dB)

    The SOX thing appears to be scaling it so amplitude values will range from -1 to 1, so a sample value of 32767 will equal 1.0, 16383 will equal 0.5, -16383 = -0.5, and so on. So 0.859971 (the peak value listed) x 32768 = 28179.53 (the peak value listed by Audition).
    I don't know what the one using the -RMS switch is doing...

    The 32768 'magic' value is 2^15, meaning it is how many levels can be represented by 15 bits. A 15 bit value can range from 0 to 32767.
    The 16th bit is used to indicate the sign of the value; if it's 1, the sample is a negative value, if it's 0 the sample is positive.

    The RMS value gives you roughly what it would read on a VU meter.


    • #3
      Thanks, Philbo! That gets me started. I wanted to use SoX to coerce the same kind of peak and RMS values that I used in SF and Audition.

      SoX is a utility that you can run it in batch or shell scripts - so I wrote a tiny script that reads a directory of files and leaves a litter of little text files that I can read with another script. Most people who use the stats in SoX just use that last "Volume adjustment:" to normalize.