Audacity Forum

Hello, hope this is new - not found on search.

Using Audacity to plot spectrum.
When "Analyse Spectrum" is used, produces a nice result, but the sound levels vary depending on the sample rate chosen, and I do not understand why.

Example:
When 94 dB 1000 Hz calibrator is used on microphone, the sound level of the 1000 Hz peak is roughly the same (-10.5 dB for my preamp and microphone) whether I use the sample rate of 16384 / 8192 / 4096 / 2048 / 1024 / 512 / 256 / 128 as expected.

When I analyse a sound recording of an environmental sound, using the same microphone and preamp setting, just looking at the 1000 Hz signal for example, I get:
16384 = -47.1 dB / 8192 = -46 dB / 4096 = -43 dB / 2048 = -40 dB / 1024 = -32 dB (a spread of 15 dB)
Similarly at all frequencies, the readings are different - spread at 63 Hz is 35 dB, and spread at 16000 Hz is 12 dB.

Why are they not all the same, as are the calibrator readings?

I would assume that if the 94 dB calibrator gives -10.5 dB, then I should be able to add 94 + 10.5 dB to the readings at any frequency and get a corrected dBA level for that frequency ... but the 15 dB difference between the 16384 rate and the 1024 rate puzzle me.

Help???

Thanks, Bill Palmer

If you have a complex waveform -- a square wave instead of a pure sine wave, then the analysis will vary most widely depending on the accuracy of your sweep. A square wave is made up of multiple sine waves adding and subtracting as needed. If you do a sloppy sweep, you may just get the square wave fundamental and that's it. If you do a much higher accuracy sweep (higher "size" setting) then you will see the individual sine waves and not the square wave. Those two displays will be very different.

We are warned not to try and convert spectrum analyses into other measurements. In addition, you may get different non-sweep numbers depending on which sound measurement you adopt; ANSI-VU, BBC-PPM, Digital Peak. All different ways of measuring the dB value of a musical or non-sine performance.

Koz

Thank you, yes, the waveform I am looking at is a complex one - environmental sound, either background, or as influenced by nearby man made noise sources. In some cases you can see the spectrum signature of things like bird songs and crickets ... or the man made sound, but what was puzzling me was that the same source recording, when the spectrum was plotted using a different sample rate produced similar frequency patterns (except for the low frequency of course where the higher sample rate gives a wider spectrum) - but with different amplitudes. That was what was puzzling me. I could not understand why the lower resolution spectrums seemed to produce higher amplitudes than the higher resolution spectrums.

Thanks though, I'll keep exploring - just another mystery for now.

Bill Palmer

Think of taking the level in each frequency band, converting to power then adding all the bands. For the same sound, if there are fewer bands there must be more power in each band.

-- Bill