I am new to matlab and I'm trying to filter the noise out of the audio. I tried some filters but can't event get close to lessening the noise.

[originalAudio, Fs] = audioread('original.wav');

[noisyAudio, ~] = audioread('noisy.wav');

t = (0:length(originalAudio)-1)/Fs;

figure;

subplot(2,1,1);

plot(t, originalAudio);

xlabel('Time (s)');

ylabel('Amplitude');

title('Original Audio');

subplot(2,1,2);

plot(t, noisyAudio);

xlabel('Time (s)');

ylabel('Amplitude');

title('Noisy Audio');

N = length(originalAudio);

f = Fs*(0:(N/2))/N;

originalSpec = fft(originalAudio);

noisySpec = fft(noisyAudio);

figure;

subplot(2,1,1);

plot(f, abs(originalSpec(1:N/2+1)));

xlabel('Frequency (Hz)');

ylabel('Magnitude');

title('Original Audio Spectrum');

subplot(2,1,2);

plot(f, abs(noisySpec(1:N/2+1)));

xlabel('Frequency (Hz)');

ylabel('Magnitude');

title('Noisy Audio Spectrum');

fc1 = 2000;

fc2 = 500;

[b1, a1] = butter(6, fc1/(Fs/2), 'low');

[b2, a2] = butter(6, fc2/(Fs/2), 'low');

filteredAudio1 = filter(b1, a1, noisyAudio);

filteredAudio2 = filter(b2, a2, filteredAudio1);

figure;

plot(t, filteredAudio2);

xlabel('Time (s)');

ylabel('Amplitude');

title('Filtered Audio');

filteredSpec2 = fft(filteredAudio2);

figure;

plot(f, abs(filteredSpec2(1:N/2+1)));

xlabel('Frequency (Hz)');

ylabel('Magnitude');

title('Filtered Audio Spectrum');

figure;

subplot(2,1,1);

plot(t, originalAudio);

xlabel('Time (s)');

ylabel('Amplitude');

title('Original Audio');

subplot(2,1,2);

plot(t, filteredAudio2);

xlabel('Time (s)');

ylabel('Amplitude');

title('Filtered Audio (Closest to Original)');

sound(filteredAudio2, Fs);

I have the noisy file and the original file that is attached: I need to make the noisy file as close to the original as possible.

[Sam Chak]

Hi @Godwin Emmanuel, you can actually attach a file here but only these formats are supported:

.bmp, .csv, .fig, .gif, .jpg, .jpeg, .m, .mat, .mdl, .mlapp, .mlx, .sfx, .slx, .sbproj, .pdf, .png, .txt, .xls, .xlsx, .zip

Click the paperclip icon .

Also click the indentation icon to copy/paste your MATLAB code that shows your filter design approach.

[Godwin Emmanuel]

The file is the audio file how do I attach them so people can hear it

[Star Strider]

Use the ‘paperclip’ (just to the right of the icon) and follow the instructions. If it is a non-standard format or extension, use the zip function to enclose it in a .zip file and then post that.

[Godwin Emmanuel]

Thank you!

[Star Strider]

My pleasure!

Answer 1

[Star Strider]

If it contains broadband noise, a frequency-selective filter will not be able to eliminate the noise. In that event, experiment with either wavelet denoising or use the sgolayfilt function. The Savitzky-Golay filter is usually effective at eliminating broadband noise, although it requires experimentation. I usually use a 3-degree polynomial and then change the ‘framelen’ value until I get the result I want.

EDIT — (28 Apr 2024 at 15:15)

The ‘noisy’ file definitely has broadband noise, however it may not be possible to filter it back to the original file.

Uz = unzip('Wavfiles.zip')

Uz = 1x3 cell array
    {'Wavfiles/'}    {'Wavfiles/noisy.wav'}    {'Wavfiles/original.wav'}

figure
tiledlayout(4,1)
for k = 1:2
    fn = Uz{k+1}
    [y{k},Fs(k)] = audioread(fn);
    ysize = size(y{k},1);
    t{k} = linspace(0, ysize-1, ysize).'/Fs(k);
    nexttile
    plot(t{k}, y{k})
    grid
    xlabel('Time')
    ylabel('Amplitude')
    title(fn)
    [FTs1{k},Fv{k}] = FFT1(y{k},t{k});
    nexttile
    plot(Fv{k}, abs(FTs1{k}))
    grid
    xlabel('Frequency')
    ylabel('Magnitude')
    title(fn)
end

fn = 'Wavfiles/noisy.wav'

fn = 'Wavfiles/original.wav'

y1_filt = sgolayfilt(y{1}, 3, 201);

figure
tiledlayout(2,1)
nexttile
plot(t{1}, y{1})
grid
xlabel('Time')
ylabel('Amplitude')
title('Unfiltered')
nexttile
plot(t{2}, y1_filt)
grid
xlabel('Time')
ylabel('Amplitude')
title('Filtered')

function [FTs1,Fv] = FFT1(s,t)
    t = t(:);
    L = numel(t);
    if size(s,2) == L
        s = s.';
    end
    Fs = 1/mean(diff(t));
    Fn = Fs/2;
    NFFT = 2^nextpow2(L);
    FTs = fft((s - mean(s)) .* hann(L).*ones(1,size(s,2)), NFFT)/sum(hann(L));
    Fv = Fs*(0:(NFFT/2))/NFFT;
%     Fv = linspace(0, 1, NFFT/2+1)*Fn;
    Iv = 1:numel(Fv);
    FTs1 = FTs(Iv,:);
end

I will consider other possibilities, however this is likely the best I can do with this file. The noise completely swamps the signal.

.

[Godwin Emmanuel]

It is broadband noise. How do I Sir use those function in my code?

[Godwin Emmanuel]

Unable to perform assignment because brace indexing is not supported for variables of this type.

When I run your code Sir I am having this error. Also how can I make a wav file of the filtered version?

[Godwin Emmanuel]

I have run the code and collected the filtered wav file Sir@Star Strider but I can't hear anything on the audio.

[Star Strider]

I am not certain where the ‘brace indexing’ error gets thrown. If you do not need to use the .zip file, then you may need to change my code to match your files. (I have to use it here because of the way the files are posted. Make appropriate changes to my code to work with your data.)

To save the filtered data, use the audiowrite function using the .wav file extension (suffix).

[Godwin Emmanuel]

This is the filtered data and is this what it should sound base on the plot Sir @Star Strider. Or I may have cause something that made it sound like this?

[Star Strider]

It is not possible to eliminate a significant amount of noise from that file. I doubt that there is any way to recover the original information from it. The noise simply overwhelms the original information.

I used several different values for ‘framelen’ from 11 to 501, and did not get a decent result from any of them.

[Godwin Emmanuel]

So this is the best I will get Sir @Star Strider if it is. I thank you very much! and I will pin this as the answer

[Godwin Emmanuel]

Btw I edited the file to this

% Unzip the files

try

Uz = unzip('Wavfiles.zip');

catch

error('Failed to unzip the file. Make sure "Wavfiles.zip" exists in the current directory.');

end

% Check if any files were extracted

if isempty(Uz)

error('No files were extracted from the ZIP archive.');

end

% Initialize variables

y = cell(1, 2);

Fs = zeros(1, 2);

t = cell(1, 2);

FTs1 = cell(1, 2);

Fv = cell(1, 2);

% Plotting and analyzing audio files

for k = 1:min(2, numel(Uz)-1)

try

fn = Uz{k+1};

[y{k}, Fs(k)] = audioread(fn);

ysize = size(y{k}, 1);

t{k} = linspace(0, ysize-1, ysize) / Fs(k);

figure;

tiledlayout(2, 1);

% Plot original waveform

nexttile;

plot(t{k}, y{k});

grid on;

xlabel('Time');

ylabel('Amplitude');

title(['Original Audio - ', fn]);

% Calculate and plot frequency spectrum

[FTs1{k}, Fv{k}] = FFT1(y{k}, t{k});

nexttile;

plot(Fv{k}, abs(FTs1{k}));

grid on;

xlabel('Frequency');

ylabel('Magnitude');

title(['Frequency Spectrum - ', fn]);

catch

warning(['Error processing file ', fn]);

end

end

% Smoothing the first audio file

try

y1_filt = sgolayfilt(y{1}, 3, 201);

% Normalize the filtered audio to [-1, 1] range

y1_filt_normalized = y1_filt / max(abs(y1_filt));

filtered_filename = 'filtered_audio.wav'; % Define the filename for the filtered audio

audiowrite(filtered_filename, y1_filt_normalized, Fs(1)); % Save the filtered audio

catch

warning('Error smoothing the first audio file.');

y1_filt_normalized = [];

filtered_filename = '';

end

% Plot unfiltered and filtered audio

figure;

tiledlayout(2, 1);

try

% Plot unfiltered audio

nexttile;

plot(t{1}, y{1});

grid on;

xlabel('Time');

ylabel('Amplitude');

title('Unfiltered Audio');

% Plot filtered audio

nexttile;

if ~isempty(y1_filt_normalized)

plot(t{1}, y1_filt_normalized);

grid on;

xlabel('Time');

ylabel('Amplitude');

title('Filtered Audio');

else

warning('Filtered audio is empty.');

end

catch

warning('Error plotting audio.');

end

function [FTs1, Fv] = FFT1(s, t)

try

t = t(:);

L = numel(t);

if size(s, 2) == L

s = s.';

end

Fs = 1 / mean(diff(t));

Fn = Fs / 2;

NFFT = 2^nextpow2(L);

FTs = fft((s - mean(s)) .* hann(L), NFFT) / sum(hann(L));

Fv = Fs * (0:(NFFT/2)) / NFFT;

Iv = 1:numel(Fv);

FTs1 = FTs(Iv,:);

catch

error('Error in FFT calculation.');

end

end

[Star Strider]

My pleasure!

That appears to be appropriate.

It is unfortunate that the noise in the file corrupted it so thoroughly. That file resisted everything I could think of to reduce or elilminate the noise.

[Sam Chak]

In practice, what reliable options do we have to attenuate the effect of broadband noise before it corrupts the original audio during measurement? Restoring the original audio from the noise-corrupted audio seems like a challenging task, although I acknowledge that noise-corrupted audio is still audible to humans to a certain extent.

If we train an AI Voice Classifier using the original audio, do you think it would be capable of improving the intelligibility of the audio by reconstructing certain parts of the signal?

[Star Strider]

@Godwin Emmanuel — Thank you!

@Sam Chak — That could work. I don’t have sufficient experience with A.I. audio processing to attempt it. I am not certain how to approach this problem from that perspective, however I would definitely like to find out how to do it.

Answer 2

[Image Analyst]

Did you try the Data Cleaner app on the Apps tab of the tool ribbon?

Can you upload a screenshot of your noisy original data, and of the denoised signal that you say is not denoised enough?

If you have any more questions, then attach your screenshots and code with the paperclip icon after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

[Godwin Emmanuel]

I uploaded an image showing the datas

Answer 3

[Ivan]

are there other methods that can be used?