Spectrum Comparisons

I am having an issue that I am hoping Audcity can resolve for me. I am wanting to export the freq/db levels from the spectrum analyzer after recording a sample of audio. Then I want to do the same thing a second time and compare the two spectrum data export files.

I was expecting to see the exact same data (at least at the start of the comparison) but strangely me DB levels are off slightly between the two files.

The were recorded on the same system about 45 seconds apart and of course same audio.

Is my basic expectation that these exports would be the same wrong, or what is it I am not understanding. It probably goes without saying I am fairly new to audio analysis.

Can you help me?

Audacity automatically adjusts the x/y axes to fit the data. I agree that this is not useful when you want to compare spectra.

The next version of Audacity (2.1.1) goes some way to improving this as it allows you to zoom the Y axis (dB). It does not completely resolve the issue but it is a step in the right direction which hopefully can be improved in later versions.

Probably the best way at present is to export the data from “Plot Spectrum”, then use third party software to plot the graph.

There are a number of “feature request” on the Audacity wiki that relate to this. Do you wish to “vote” for any of these: http://wiki.audacityteam.org/wiki/Feature_Requests#Plot_Spectrum (to “vote”, just say which you want and I can update the vote count).
(note that “voting” does not mean that the feature will be developed, but it gives the Audacity team an idea about what features users are interested in.)

Thanks for the reply. What I do not understand is how the x/y axes can be plotted differently when it is the exact same audio file that is being played in both sides of the comparison. The frequencies match exactly in the comparisons but the decibels do not. How is that possible?

A possible explanation:

It is often a question of proper alignment.
Spectrograms are calculated from (overlapping, windowed) frames.

Thus, it is important that the audio fits into the “grid”.
A numerical example:
Our sequence is
1 2 3 0 0 1 2 3
Frame and hop size = 2, operation = sum:
Frame 1 = 3 (1+2)
Frame 2 = 3 (3+0)
Frame 3 = 1 (0+1)
Frame 4 = 5 (2+3)

As you can see, frames 1-2 and 3-4 can’t be compared although they have the same numbers and their overall sum is the same (namely 6).

Another example:
The sequence is
1 1 1 1 1 1 1 1

Frame and hop size = 3, operation = add and average frames (as in “Plot spectrum”)

Frame 1 = 1 1 1
Frame 2 = 2 2 2 (frame 1 is added)
Frame 3 = 3 3 2 (we have to add a zero to the last frame)
divided by number of frames = 3:
result = 1 1 0.666

The exact same happens to the spectrogram view and the plot spectrum effect, the operation is just different (a Fourier transform instead of a summation).

A proper comparison has therefore to be made with multiples of the chosen window sizes in mind.


Excellent explanation, thanks.