There isn't any calculating difference at all - it is an essentially identical exercise in scaling. If it wasn't, there would have been complaints about any differences years ago, and there haven't been. And that's going back to Syntrillium days, and Cool Edit!
There is one area where a little caution should be exercised, though. Amplifying or reducing the level of a 16-bit integer file and saving the results will give you a file with an altered bit depth permanently. This is far worse with reducing the level; every 6dB drop in the peak level will result in a 1-bit loss. So if you have a file that was peaking at 0dB and you reduced its level to -18dB and saved it, you would have effectively saved it as a 13-bit file. If you reopen it and decide to normalize it back to 0dB, it will still have an effective bit depth of 13dB - basically the noise floor will come up with it.
None of this happens with 32-bit Floating Point files - they can be scaled up and down, saved, and then re-scaled without loss. This is because it isn't the signal information that is altered when you amplify or normalize - each sample is stored permanently as a value between -1 and +1. What's altered is the scaling of this, and as that's just a number, you can alter it at any time without any penalty at all. The format for the file is that the signal is a 23-bit mantissa, there is one polarity bit, and an 8-bit scaling exponent. So the audio is stored primarily as 24-bit, which gives a dynamic range that exceeds the capability of any hardware to use - the Laws of Physics see to that.
It does of course follow that the computation done to an integer signal for amplification purposes is completely different to that done for a Floating Point 32-bit one. But this has all been very carefully worked out years ago, and it is essentially error-free - with the provisos mentioned for integer-based calculations.
1 Correct answer
2
Replies
AdChoices