Test tone sloping down

[Pilot]

Thank you for this great app!

I am using a Pixel 8 Pro. The test tone slopes down as shown. Does this mean my phone always performs this type of adjustment (despite having chosen "unprocessed" as the source in the settings)?

[Pilot]

Updated findings, the "sloping down" occurs when choosing "logarithmic sweep" as the test tone. A linear sweep or white noise are flat:

  

The settings for the two tests above are

 

However, introducing "decimations" will cause the white noise and linear sweep tone to slope upwards:

 

Reason I am looking for a flat test tone: I want to use Spectroid to estimate the in-room response of my stereo speakers and subwoofer. I realize there are limitations to what the phone can do in this regard, however, as a tool for comparisons between audio settings and positioning, Spectroid seems adequate. I want to make sure however that I am measuring as well as it can. From what I've seen so far, the best settings for this purpose seem to be those pictured above, (max FFT, no decimations, max interval, no smoothing).

Thanks again!

[Carl Reinke]

Seems like you got it figured out. :)

Cascaded decimations are a cheap trick to produce a spectrum that's probably nicer to look at but maybe not so great for analysis since the frequency bins aren't all the same width.

The downward slope of the logarithmic frequency sweep is probably just an artifact of the speed of the sweep.  If I made the sweep slow enough, I suspect it would not occur.

--Carl

[Pilot]

Thanks!

Since it's a test tone artifact and doesn't affect the measurements, there is no need to resolve it further; 

but, out of curiosity, the way I understand it is that the log sweep begins going through each Hz too fast, for the phone's CPU or the "loopback" code perhaps, at some point. But when I change the "desired transform interval" (which I understand as the app's "frame rate"), from 10ms to 100ms, the slope is the same.

As an aside, I've been reading up a bit on FFT and I understand that for my case of doing a smooth and slow sweep to measure in-room frequency response, the window I should be using is "Flat top", to minimize spectral leakage?

[Carl Reinke]

For the frequency sweep test signals, the frequency being generated changes for every sample produced.  When it's time to do a transform, it uses the latest samples (however many it needs) as input.

If the frequency is changing slowly enough, the power in that chunk of samples will fall into one or two frequency bins.  As the generated frequency changes more quickly, it'll be spread across more frequency bins, so the amount of power in any one of those bins will be less.

For a given sample rate, the number of samples in the chunk corresponds to the duration of audio being analyzed, during which the frequency being generated changed by some amount.  (For a linear sweep the frequency change is constant; for logarithmic it increases as frequency increases.)  The interval between transforms doesn't change that duration of audio being transformed and so doesn't affect the amount of spreading.  (But if the interval between transforms is too long, there will be audio that didn't get analyzed.  If I were to start over, Spectroid wouldn't have the transform interval option and would just aim for the interval necessary to not miss any audio.)

Flat-top provides the best amplitude accuracy when analyzing a tone.  Rectangular provides the best frequency resolution.  Beyond that, choosing a window function is more engineering than science -- i.e., pick whatever seems to work well.

--Carl

[Pilot]

Many thanks for that explanation (and now excuse me while I go read it again 😄).

Spectroid is a great app, appreciate all the work you've put in it.