Spectral Feature Block

Help
#1

Hi, I was trying to understand how the output of the spectral block is affected by changes in the FFT parameters (FFT length and cut-off). However, the number of extracted features in the Studio doesn’t seem to match the expected outcome I had in mind. Specifically, there were always 2 extra features . I observed in the git repo (processing-blocks/spectral_analysis/dsp.py at master · edgeimpulse/processing-blocks · GitHub) and the python example (processing-blocks/run_spectral_analysis_via_python.py at master · edgeimpulse/processing-blocks · GitHub) that there are two spectrum statistical (spec_stats) features extracted (spectral skew and kurtosis ) when the implementation version is 4 ( as it is in the python example).

Are those features indeed extracted? I could not find any information about spectrum statistical features to confirm this.

Thank you!

#2

Hi @Chrysanthi

Let me ask one of the DSP teams opinion here, spec_stats I am not so sure about but they should be available if they are mentioned in the sample code, even if not displayed in the studio UI. @yanedge @AlexE

Best

Eoin

#3

Checking the docs here should also confirm the features that are available:

  • Statistical features

    • RMS

    *** Skewness**

** * Kurtosis**

  • Spectral features

    • Maximum value from FFT frames for each bin that was not filtered out

Spectral features | Edge Impulse Documentation

Best

Eoin