trouble making a firfilter to remove a tone out of a wavefile

I am trying to write a matlab code to design a firfilter to filter a tone out of a wav.file but I am having trouble getting the filter to work this code below is supposed to determine the frequencies of the audio in the wav.file and filter the tone out of the sunshinesquare.wav file. however when I go to run the program matlab tells me there is an error with the notchfilter line but I can't figure out what is wrong with it any suggestions/help is greatly appreciated.

My code is below

[x, fs] = audioread('SunshineSquare.wav'); 

X = fft(x); 
f = linspace(0, fs/2, length(X)/2);
[~, f0] = max(abs(X(1:length(X)/2))); 

notchFilter = fdesign.notch('N,Fc', 100, f0, fs); 
b = firpm(notchFilter); 

y = filter(b, 1, x); 
sound(x, fs);

[Walter Roberson]

Cross-reference https://www.mathworks.com/matlabcentral/answers/2053427-trouble-making-a-notchfilter-to-remove-a-tone-out-of-a-wavefile?s_tid=srchtitle

[Walter Roberson]

Error using fdesign.abstractpeaknotch/set.Specification

Expected Specification to match one of these values:

'N,F0,Q', 'N,F0,Q,Ap', 'N,F0,Q,Ast', 'N,F0,Q,Ap,Ast', 'N,F0,BW', 'N,F0,BW,Ap', 'N,F0,BW,Ast', 'N,F0,BW,Ap,Ast'

The input, 'N,Fc', did not match any of the valid values.

[Chunru]

issues with making a matlab firfilter - MATLAB Answers - MATLAB Central (mathworks.com)

Answer 1

[Walter Roberson]

The error is correct, in that the documentation lists specific filter specifications, none of which include Fc

https://www.mathworks.com/help/dsp/ref/fdesign.notch.html#d126e434241

I do not promise that I calculated f0 correctly.

[x, fs] = audioread('SunshineSquare.wav'); 
X = fft(x); 
L = floor(length(X)/2);
f = linspace(0, fs/2, L);
[~, f0idx] = max(abs(X(1:L))); 

f0 = (f0idx-1)/L;
Q = 80;
notchFilter = fdesign.notch('N,F0,Q', 100, f0, Q); 
Hd = design(notchFilter, 'SystemObject', true);
y = Hd(x);
sound(y, fs)

[Jacob]

Okay I tried the changes and no matter what I do there is still a tone in the audio. Not sure what I am doing wrong there.

[Walter Roberson]

Depending how long your sound is relative to fs, each bin of the fft might be fairly narrow.

Example code:

%[x, fs] = audioread('SunshineSquare.wav'); 
load handel
x = y; fs = Fs;

X = fft(x); 
L = floor(length(X)/2);
f = linspace(0, fs/2, L);
[~, f0idx] = max(abs(X(1:L))); 

f0 = (f0idx-1)/L;
Q = 2.5;
notchFilter = fdesign.notch('N,F0,Q', 6, f0, Q); 
Hd = design(notchFilter, 'SystemObject', true);
y = Hd(x);
%sound(y, fs)

tiledlayout('flow');
nexttile
plot(f, abs(X(1:L))); title('original spectrum');
nexttile
Y = fft(y);
plot(f, abs(Y(1:L))); title('filtered spectrum');

If you sound(y,fs) then it is difficult to hear the difference, but if you look at the plot of the fft, cclearly the largest peak has been filtered out.

[Jacob]

Is there a good way to use sound(y,fs) to hear the difference? I need the plot but I also need to demonstrate with a wav file too.

[Les Beckham]

To hear a significant difference, you probably need to filter out the harmonic at 1200 Hz as well. You might also want to try making your notch a bit wider.

[Jacob]

Okay so would i just do the same thing like this then??

f0 = (f0idx-1)/L;

Q = 5.5;

[Les Beckham]

I'm not sure if making Q bigger makes the notch wider or narrower. You'll have to study the fdesign.notch documentation, or experiment.

To filter out the harmonic you could just repeat the design/filtering process starting with finding the max of the fft again after applying the filter the first time.

[Walter Roberson]

%[x, fs] = audioread('SunshineSquare.wav'); 
load handel
x = y; fs = Fs;

X = fft(x); 
L = floor(length(X)/2);
f = linspace(0, fs/2, L);
[~, f0idx] = max(abs(X(1:L))); 

f0 = (f0idx-1)/L;
Q = 2.5;
notchFilter1 = fdesign.notch('N,F0,Q', 6, f0, Q); 
notchFilter2 = fdesign.notch('N,F0,Q', 6, f0*2, Q);

Hd1 = design(notchFilter1, 'SystemObject', true);
Hd2 = design(notchFilter2, 'SystemObject', true);
y = Hd2(Hd1(x));
%sound(y, fs)

tiledlayout('flow');
nexttile
plot(f, abs(X(1:L))); title('original spectrum');
nexttile
Y = fft(y);
plot(f, abs(Y(1:L))); title('filtered spectrum');

[Walter Roberson]

Higher Q gets a narrower notch, so a wider notch needs a lower Q

%[x, fs] = audioread('SunshineSquare.wav'); 
load handel
x = y; fs = Fs;

X = fft(x); 
L = floor(length(X)/2);
f = linspace(0, fs/2, L);
[~, f0idx] = max(abs(X(1:L))); 

f0 = (f0idx-1)/L;
Q = 2.5;
notchFilter1 = fdesign.notch('N,F0,Q', 6, f0, Q); 
notchFilter2 = fdesign.notch('N,F0,Q', 6, f0, Q/10);

Hd1 = design(notchFilter1, 'SystemObject', true);
Hd2 = design(notchFilter2, 'SystemObject', true);
y1 = Hd1(x);
y2 = Hd2(x);
%sound(y, fs)

tiledlayout('flow');
nexttile
plot(f, abs(X(1:L))); title('original spectrum');
nexttile
Y1 = fft(y1);
plot(f, abs(Y1(1:L))); title('filtered spectrum Q 2.5');
nexttile
Y2 = fft(y2);
plot(f, abs(Y2(1:L))); title('filtered spectrum Q .25');

[Jacob]

Okay so still unable to edit out the tone from my file I tried adding another row of these

X = fft(x); 
L = floor(length(X)/2);
f = linspace(0, fs/2, L);
[~, f0idx] = max(abs(X(1:L))); 
f0 = (f0idx-1)/L;
Q = 2.5;
notchFilter1 = fdesign.notch('N,F0,Q', 6, f0, Q); 
notchFilter2 = fdesign.notch('N,F0,Q', 6, f0*2, Q);
Hd1 = design(notchFilter1, 'SystemObject', true);
Hd2 = design(notchFilter2, 'SystemObject', true);
y = Hd2(Hd1(x));

But that didn't work Do I need to bassically do a double of the program to get it work?

[Walter Roberson]

When you plot the fft like I showed,

Y = fft(y);
plot(f, abs(Y(1:L))); title('filtered spectrum');

then are the peaks visually removed?

That is, is it a difficulty with not filtering at all? Is it a difficulty with the notching being smaller than desireable? Is it an issue that the transformation is not making sufficient audible change?

[Jacob]

Yes like you said when I run this program the peaks are removed but there is no audible change it sounds the same as not filtered at all.

[Walter Roberson]

In my testing with handel, the result of the filtering was sound that was quieter and less "bright", but still very recognizably the original song.

[Jacob]

Okay still doesn't work is there anything else to try because I do need the audio file to be clean

[Walter Roberson]

Can you zip the wav file and attach the zip?

[Jacob]

Okay I'll get that uploaded it's only about 10 seconds.

[Jacob]

here you go

[Walter Roberson]

The audiofile has a primary frequency and two further peaks at 2 and 3 times the primary. Removing those three leaves only low level information.

X = fft(x); 
L = floor(length(X)/2);
f = linspace(0, fs/2, L);
[~, f0idx] = max(abs(X(1:L))); 
f0 = (f0idx-1)/L;
Q = 2.5;
notchFilter1 = fdesign.notch('N,F0,Q', 6, f0, Q); 
notchFilter2 = fdesign.notch('N,F0,Q', 6, f0*2, Q);
notchFilter3 = fdesign.notch('N,F0,Q', 6, f0*3, Q);
Hd1 = design(notchFilter1, 'SystemObject', true);
Hd2 = design(notchFilter2, 'SystemObject', true);
Hd3 = design(notchFilter3, 'SystemObject', true);

y = Hd3(Hd2(Hd1(x)));

tiledlayout('flow');
nexttile
plot(f, abs(X(1:L))); title('original spectrum');
nexttile
Y = fft(y);
plot(f, abs(Y1(1:L))); title('filtered spectrum');

[Jacob]

Awesome Thank you! It works and I don't hear the tone anymore which is what I needed! One more thing the end of the file is still a little noisy but I can hear the file is there a way to clean the file up or is it what it is? I know there is only so much that can be done cleaning up an audio file so I am happy with what I have if not.