How do i create a comparison loop of two signals based on their spectral centroid value, with a sliding filter dependant on outcome to balance the amended spectrum

this is currently the code i have where the loop doesnt really do what i'd like, but im not sure how is best to loop it correctly. what i need is to compare spectral centroid (SC) A against spectral centroid B which has been run through a filter the reduce the fundamental frequencies. where SC B is now above SC A then a shelf filter should be run on spectrum B reducing the higher frequencies by 1db to reduce the SC down to the same (or withing a small margin of error) as SC A. ideally the shelf filter would start at the highest frequency and then slide downwards, until the SC are the same. would anybody know a reasonable way of doing this?

%% read audio file

[pop,fs] = audioread("Pop Base.wav");

%% SC of Base sound

centroidA = spectralCentroid(pop,fs);

ScA1 = mean(centroidA);

ScA = ScA1(:,1)

%% Fundamental Reduction

gain1 = -6;

slope1 = 2;

Fc1 = 50;

lowFilt = shelvingFilter(gain1,slope1,Fc1,"lowpass");

%visualize(lowFilt)

popHP = lowFilt(pop);

%outcome called popHP

%% SC of ammended Sound

centroidB = spectralCentroid(popHP,fs);

ScB1 = mean(centroidB);

ScB = ScB1(:,1)

%% loop for spectral sweeping

if ScB > ScA

% apply LP shelf filter

gain2 = -1;

slope2 = 2;

Fc2 = ScA;

highFilt = shelvingFilter(gain2,slope2,Fc2,"highpass");

%visualize(highFilt)

popLP = highFilt(popHP);

centroidB = spectralCentroid(popLP,fs);

ScB1 = mean(centroidB);

ScB = ScB1(:,1)

end

audiowrite("C:\Users\peter.wheeler\OneDrive - Acoustic Consultants\" + ...

"Documents\MATLAB\poptest.wav",popLP,fs)

Answer 1

[Brian Hemmat]

Hi Peter, I'm not sure I fully understand your problem statement. Here's some sample code for what I think you're asking about, but please let me know if I'm misinterpretting.

% Read audio file and convert to mono
[x,fs] = audioread("WashingMachine-16-44p1-stereo-10secs.wav");
x = mean(x,2);

% Calculate mean spectral centroid
% This is the target spectral centroid
SC_A = spectralCentroid(x,fs);
SC_A = mean(SC_A,"all",'omitmissing');

% Filter signal
gain1 = -6;
slope1 = 2;
Fc1 = 50;
lowFilt = shelvingFilter(gain1,slope1,Fc1,"lowpass",SampleRate=fs);
visualize(lowFilt)
popHP = lowFilt(x);

% Calculate mean spectral centroid of filtered signal.
% You want to apply a high-pass filter so that the spectral centroid of
% the filtered signal is similar to the spectral centroid of the original
% signal (SC_A)
SC_B = spectralCentroid(popHP,fs);
SC_B = mean(SC_B,"all","omitmissing");

% Design initial filter. Start at the highest frequency
gain2 = -1;
slope2 = 2;
Fc2 = fs/2;
highFilt = shelvingFilter(gain2,slope2,Fc2,"highpass",SampleRate=fs);
visualize(highFilt)

% Define parameters for the sweep.
threshold = 1;
frequencystep = 400;
frequencyStepReduction = 2;

% Run the sweep until threshold met
while abs(SC_B - SC_A) > threshold

    % Apply 
    popLP = highFilt(popHP);
    SC_B = spectralCentroid(popLP,fs);
    SC_B = mean(SC_B,"all","omitmissing");

    if SC_B < SC_A
        % Overshot, reduce frequency step
        SC_B = inf;
        highFilt.CutoffFrequency = highFilt.CutoffFrequency + frequencystep; % return to before overshoot
        frequencystep = frequencystep/frequencyStepReduction;
    else
        highFilt.CutoffFrequency = highFilt.CutoffFrequency - frequencystep;
    end
    
end

% Look at the final parameters of the high-shelf filter.
highFilt.CutoffFrequency = highFilt.CutoffFrequency - frequencystep

% Compare spectral centroids of the target, filtered, and re-filtered signals
SC_A = mean(spectralCentroid(x,fs),"all","omitmissing")
SC_B = mean(spectralCentroid(popHP,fs),"all","omitmissing")
SC_C = mean(spectralCentroid(popLP,fs),"all","omitmissing")

% Listen to the results.
soundsc(x,fs),pause(size(x,1)/fs+0.5)
soundsc(popHP,fs),pause(size(x,1)/fs+0.5)
soundsc(popLP,fs),pause(size(x,1)/fs+0.5)

[Peter Wheeler]

Hi Brian,

thank you for this. for the most part i think this covers everything. i'll just do a little bit of tweaking for final outcome. i appreciate the help