ai_audio_cleaning.m

% Load clean audio
[cleanAudio, Fs] = audioread('non_noise.wav');
cleanAudio = cleanAudio(:, 1); % Force to mono / first channel if stereo

% Load noisy audio
[noisyAudio, ~] = audioread('noisy_music.wav');
noisyAudio = noisyAudio(:, 1); % Force to mono / first channel if stereo

% Match lengths strictly
minLength = min(length(cleanAudio), length(noisyAudio));
cleanAudio = cleanAudio(1:minLength);
noisyAudio = noisyAudio(1:minLength);

% Ensure they are vertical column vectors
cleanAudio = cleanAudio(:);
noisyAudio = noisyAudio(:);

% Pre-filter noisy audio (CRITICAL: Match workshop live-demo pipeline)
noisyAudio = lowpass(noisyAudio, 4000, Fs);

% Create features for AI (Must be exact same length column vectors)
X = [
    noisyAudio, ...
    movmean(noisyAudio, 5), ...
    movmean(noisyAudio, 20), ...
    movmean(noisyAudio, 50), ...
    movmean(noisyAudio, 100), ...
    movmean(noisyAudio, 200)
];

% Normalize the feature matrix to stabilize fitlm tracking
X = normalize(X);

% Train Linear Regression Model
model = fitlm(X, cleanAudio);  % This will now execute perfectly

% Predict cleaned audio
predictedAudio = predict(model, X);  % asks model to predict cleaner audio based on what it learned

% Final smoothing step
predictedAudio = movmean(predictedAudio, 8);  % applies smoothing after predicting audio

% Normalize output audio
predictedAudio = predictedAudio / max(abs(predictedAudio));

% Save audio
audiowrite('ai_filtered.wav', predictedAudio, Fs);  % saves audio to another file name

% Play result
disp('Playing AI Cleaned Audio...')
sound(predictedAudio, Fs)