|
Computer Audio Asylum Music servers and other computer based digital audio technologies. |
|
|
|
Computer Audio Asylum Music servers and other computer based digital audio technologies. |
|
In Reply to: RE: TESTS: Round-up of Windows and Mac Audio Players. posted by Ugly on June 17, 2013 at 07:13:35:
The process gain in analysis software is very easy to measure. Start with a file of white noise at a moderately high level. (You want it to be high enough that it is way above the noise floor of the numerical calculations themselves but low enough that no peaks ever clip.)
Now you average over a long time period. You want the noise file to be long enough that examining the spectrum at a point in the middle of the file doesn't change significantly as you move the point you have selected. (You will see end effects if you get too near the start or end of the file. If you aren't seeing these then you should continue to experiment until you do. Hitting an end of the file will change the window size, since it will cause the software to see lots of zero samples that aren't in the file or, worse, provoke a bug of some sort.)
You will now see the noise floor in the plotted graph. This will vary slightly as you move the sample point around in the middle of the file, particularly where you use a very high order FFT (e.g. 65K). This is normal and happens because each bin samples only a small amount of energy and the noise is (supposedly) random.
Your software will probably also be capable of computing some kind of an RMS average. You can look at this value. Here you want to average over a long interval so the results are stable. (Note that different software may give different RMS averages even though they both are doing an unweighted computation. That is because the idiots at the AES standardized on a zero level that corresponds to a sine wave, rather than a square wave. This is basic math and electrical engineering. So you may get results that differ by 3 dB.)
The process gain is just the difference you see between the single number calculated by the averaging program and the level of the noise floor read off the scale of the graphical spectrum plot.
If you try and calibrate the graphical scale by using test tones at known levels you may also observe funny effects. If a tone is in the dead center of an FFT bin it may plot at a different peak level than if the tone is straddling two bins and most of the energy is split between the two bins. In addition to changing with frequency this effect will also vary with the FFT size and window. This is because the size of the bins changes with FFT size and the shape of the bins changes with the window filter. (The window attempts to weigh points at the center greater than points at the edge to smooth out the effects of the last points measured. This is called the Gibbs effect.)
I generally do this kind of analysis with Soundforge. It is a good tool for playing around and understanding in a practical sense what is going on. Having a theoretical analysis of Fourier transforms, FFT algorithms, RMS averages, etc., is pretty close to useless when using tools for actual work. IMO it is much better, indeed essential, to have hands on experience with one's tools. If you feel the need to rely on some external authority for magic numbers such as "39 dB" you should realize that you need more hands on experience with the tools involved.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
This post is made possible by the generous support of people like you and our sponsors:
Follow Ups
- RE: TESTS: Round-up of Windows and Mac Audio Players. - Tony Lauck 08:43:40 06/17/13 (1)
- RE: TESTS: Round-up of Windows and Mac Audio Players. - Thorsten 09:16:36 06/17/13 (0)
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
