You need to be a little more clear in your question.
The Audition amplitude statistics window calculates the RMS value of the selected chunk of audio. Actually it calculates the RMS value to each window sized chunk in the selected audio chunk. Then it displays the RMS values for the minimum window the maximum window and the average of all the windows in the selected chunk. There are various definitions of digital "0 dB". In Audition you can select between two of these using the "FS sine" or "FS square" setting. If you select "FS sine" then a maximum unclipped sine wave will read 0 dB RMS.
To measure an analog signal level the analog input gain will need to be measured and adjusted for, record a sine wave signal at a known level, measure the Adobe Amplitude reported level, and note the offset. As long as you don't change the input sound device gain setting, the reported level offset will remain the same.
Sorry for my unclear question. What I want to know is how the window shift. For example if I set the windows width to 50ms, after calculating the first 50ms RMS, which period will the next window capture to calculate RMS value? Will the next window start at the end of the first window or the second sample of the whole waveform?
>What I want to know is how the window shift. For example if I set the windows width to 50ms, after calculating the first 50ms RMS, which period will the next window capture to calculate RMS value?
I would have thought that the answer was
blindingly obvious from the help files, personally, unless you don't understand the concepts of a 'series', and 'range'. Have you not read this bit?
Window Width Specifies the number of milliseconds in each RMS window. A selected range contains a series of such windows, which Adobe Audition averages to calculate the Minimum RMS and Maximum RMS values. To achieve the most accurate RMS values, use wide windows for audio with a wide dynamic range, and narrow windows for audio with a narrow dynamic range.
I guess it is a fair question. I would assume the second window starts at the end of the first window, but I could see that from a pure theory point of view it might be possible to "slide the window" over one sample at a time. That is how my professor (many years ago)in digital signal processing would have suggested it be done. Of course that would generate a lot more windows and take a lot more processing time. Another less obvious question is how the final window(s) are handled, are any windows calculated using less than the window size number of samples?
If the different possible methods would make a difference to you, couldn't you think up a test file that would have a different answer based of which method is actually being used? This should allow you to figure out what Audition is actually doing. If you can't think up a test file, does it really matter to you?
>...but I could see that from a pure theory point of view it might be possible to "slide the window" over one sample at a time. That is how my professor (many years ago)in digital signal processing would have suggested it be done.
I don't think he would have! Let's take the standard sample window size of 50ms. At 44.1k that contains 2,205 samples. So there would be absolutely no point in sliding a dirty great window like that over a single sample at all, would there?
Disregarding that bit about dynamic range (which may well be BS), the basic idea is that you are calculating the square root of the mean value of the signal squared - that's what RMS stands for. Because you've squared the mean value of the wave, it's always going to be a positive value. And note that this is the mean value of a repetition of the
wave, not a sample. And we are talking about the
arithmetic average value of all of the waves in the sample window, so all of the positive and negative excursions are taken account of.
'Series' in this context means that the selected part of the wave being analysed is treated as a sequence of windows of whatever size you've chosen - sliding doesn't come into this at all. Statistically, Audition can give percentages of windows with low and high values, and calculate the average RMS value based on this.
The reason that I think the dynamic range bit might be BS, or at best completely back to front is that you'd need to make sure that you'd used short samples to catch the quietest and loudest parts to have them treated as such, rather than lumped in with a larger sum with either a higher or lower average value, which you'd get with longer samples. But I have to say that when I tested this with a range of different window sizes, it really didn't make a lot of difference.
Call me blind, because I'm reading the help file right now. If anything, it's blindingly ambiguous! I tend to be sarcastic myself, to the point of my own detriment, but I'll have to side with the OP here.
SteveG, I'm unclear as to what distinguishes a "wave" from samples, unless the statistics are based on a fully reconstructed signal. Otherwise, the order of samples within a given window would be unimportant. Also, it wouldn't be consistent with other reported values, such as peak amplitude, which is explicity defined as applying to a sample.
The help file's definition of "Average RMS Power" is "the average amplitude". On the contrary, it is RMS amplitude that equates to mean power. "RMS power" itself is an ill-defined term that, if taken literally, may be computed, but has no physical meaning. I'll assume that "mean power" is what is intended.
>I'm unclear as to what distinguishes a "wave" from samples, unless the statistics are based on a fully reconstructed signal.
They are - they have to be. A sample on its own means nothing. Since a waveform has to be continuous, the position of individual samples indicate no more than the rate at which the waveform is
changing amplitude. But...
>Otherwise, the order of samples within a given window would be unimportant. Also, it wouldn't be consistent with other reported values, such as peak amplitude, which is explicity defined as applying to a sample.
NO WAY is the the peak amplitude of a signal defined as applying to a sample. If you want to see why, have a look at
this document, which I prepared some time ago, that demonstrates very clearly why not.
This whole sampling business gets more misunderstood than just about anything else at all to do with digital audio. There was a huge amount of mis-information spread about in the early days - by people who should have known better - about 'slicing up' signals, and as a result, samples and sampling have got a name for themselves that they simply don't deserve.
>"RMS power" itself is an ill-defined term that, if taken literally, may be computed, but has no physical meaning.
Since the concept of 'power' involves putting energy into a load, it can't be computed directly by Audition anyway. I have complained before to Adobe about having any references to power anywhere in the software, because any mention at all is quite incorrect, and therefore wrong. The only thing that Audition can make any correct assertation about is relative amplitude.
Yes, you're quite right about peak amplitude. I'm well acquianted with signal reconstruction, and agree that we're up against a mountain of misinformation. I was referring to Adobe's documentation, in which they claim that the reported peak amplitude "shows the sample with the highest amplitude...". That doesn't sound right to me, either.
I'll make some short test files and try to determine what's really going on.