Feature Extraction in Time-Frequency Domain - Audio Classification

Hi, I'm fairly new to MATLAB, really appreciate if I can get some help here.

I'm doing a project on audio classification, where I have loaded labelled audio datasets into MATLAB. I am currently encountering an issue where I want to apply discrete wavelet transform on all the signals in the datasets and convert them to images, subsequently saving them in a folder according to their labels. The images will then be fed into a neural network for classification learning.

I am trying to replicate another projet where they created a function to automatically convert CWT coefficients into images but I'm having no luck trying to make it work for DWT.

Thanks in advance for helping, and if I'm missing out any details please let me know.

my code:

https://pastebin.com/nCxzFqeN

The function code for CWT

https://pastebin.com/P6bZG6T7

Answer 1

[Yash Sharma]

To convert audio signals to Discrete Wavelet Transform (DWT) images and save them according to their labels, you can modify the function you provided for CWT. Below is a concise guide and a modified function tailored for DWT.

Modified Function for DWT

This function, helperCreateImagesFromDWT, processes audio files, applies DWT, and saves the coefficients as images in labeled folders.

function helperCreateImagesFromDWT(ads, outputFolder)
    for i = 1:length(ads.Files)
        [audioIn, info] = audioread(ads.Files{i}); % Read audio file
        [cA, cD] = dwt(audioIn, 'db1'); % Apply DWT with 'db1' wavelet
        
        % Process and normalize coefficients
        coeffs = [cA; cD];
        coeffsNorm = (coeffs - min(coeffs)) / (max(coeffs) - min(coeffs));
        img = im2uint8(mat2gray(coeffsNorm)); % Convert to image
        
        % Prepare image file path
        label = char(ads.Labels(i));
        imgFileName = strcat(label, '_', num2str(i), '.png');
        imgFolderPath = fullfile(outputFolder, label);
        
        if ~exist(imgFolderPath, 'dir'), mkdir(imgFolderPath); end % Ensure folder exists
        
        imwrite(imresize(img, [224, 224]), fullfile(imgFolderPath, imgFileName)); % Save image
    end
end

Applying the Function

1. Set Up the Output Folder: Change datafolder_scallogram to a new folder name that reflects you're working with DWT images, for example, datafolder_dwtImages.
2. Call the Function: Replace the call to helperCreateRGBfromTF in your script with helperCreateImagesFromDWT, passing the audio datastore and the new output folder path.

datafolder_dwtImages = "C:\Users\leere\Documents\MATLAB\DWT_Images"; % New folder for DWT images
helperCreateImagesFromDWT(ads, datafolder_dwtImages); % Process and save DWT images

This approach applies DWT to your audio signals, normalizes and processes the coefficients to create images, and organizes these images in folders based on their labels, ready for neural network classification.