How to filtre an audio signal with low-pass filtre

Hey guy, I'm new on Matlab and I have no idea how can I do a low pass filtre to filtre my signal.

Here is my signal and I don't know how to remove that hight frequency from backgrond and to keep just that voice.

I would like to know the easier way to do this.

I was doing some research on google and I so a lot of complicate options, and some options with butter or filter or.. to use fdatool and so on... But I could not understand how to use them no matter how much we read about this.

And please don't give me some links from some sites or other matlab questions like an answer, I've alread read everything for nothing...

This is the why how I read and playback the sound after I've try some filtres.

[Y, Fs] = audioread('semnale_audio/semnal1.wav');
%filtred my signal some how...
 player = audioplayer(fitlered_signal, Fs);
 play(player);

So please guys, if someone could explain me how can I do this I would be very glad.

Answer 1

[Star Strider]

I cannot access your dropbox file. It would be much easier for us for you to use the ‘paperclip’ icon and upload it here instead.

Even in the absence of your file, it is easy to design your filter. Human voice frequencies are in the range of about 100 Hz to 6000 Hz, so a Chebyshev Type II filter to pass voice frequencies would be:

Fs = 44100;                                             % Sampling Frequency (Change If Different)
Fn = Fs/2;                                              % Nyquist Frequency
Wp = [150  5800]/Fn;                                    % Normalised Passband
Ws = [ 50  6100]/Fn;                                    % Normalised Stopband
Rp =  1;                                                % Passband Ripple (dB)
Rs = 30;                                                % Stopband Ripple (dB)
[n,Ws] = cheb2ord(Wp,Ws,Rp,Rs);                         % Chebyshev Type II Order
[b,a] = cheby2(n,Rs,Ws);                                % IIR Filter Coefficients
[SOS,G] = tf2sos(b,a);                                  % Convert To Second-Order-Section For Stability

figure(1)
freqz(SOS, 4096, Fs)                                    % Check Filter Performance

filt_sig = filtfilt(SOS,G, Your_Signal);

Change the appropriate passband and stopband frequencies depending on the frequency content of your signal.

[Image Analyst]

I'm not an expert in Signal Analysis so what are the advantages, disadvantages, and differences between Butterworth and Chebyshev? When is it appropriate to use each of them?

[Vlad Dumitru]

I don't know why you cant access that file on dropbox.. I've put it and on my google drive maybe will be fine for you now. And I've saw and this kind of solution when I've search on internet how to do this... but I don't know how is my Y, I mean when I'm read my file I'm doing something like this...

   [Y, Fs] = audioread('semnale_audio/semnal1.wav');

So how is my Y in your case? Sorry if it's sound stupid but I'm really new on matlab and I'm tring from hard to learn how it works.

[Star Strider]

@Image Analyst — The Butterworth design is good for most filtering applications where the rolloff is not critical, so there is ‘room’ in frequency space for gradual rolloffs. Butterworth filters have maximally flat passband and stopband frequency characteristics, making them good choices in most applications. Butterworth filters can be designed with high rolloffs, but they require long filters and can have stability problems.

Chebyshev filters are better for low-frequency applications because they have steep rolloffs and can be designed to eliminate baseline wander and d-c offsets in signals with significant low frequency content (such as EKGs).

@Vlad Dumitru — I was able to get it from the second site. I didn’t realise that ‘.wav’ files aren’t allowed here. I saved the original signal and sampling frequency to a ‘.mat’ file and attached it to this Comment.

Filter Design and Implementation —

Here, I abandoned both for either lowpass or bandpass filters because too many of the higher frequencies (plosives, fricatives, and such) were being lost. The ‘lost’ content made the voice unintelligible (at least to my ears). With a bit more complexity, I designed a lowpass FIR filter that incorporates two notch filters to ‘notch out’ the two pure tones in the signal.

The Code —

[S,Fs] = audioread('Vlad Dumitru semnal1.wav');
Fn = Fs/2;                                                      % Nyquist Frequency
L = size(S,1);                                                  % Signal Length

fcuts = [680 690 710 720  1190 1205  1210 1220  6000  6100];    % Frequency Vector
mags =   [1 0 1 0 1 0];                                         % Magnitude (Defines Passbands & Stopbands)
devs = [0.05  0.01  0.05  0.01 0.05  0.01];                     % Allowable Deviations
[n,Wn,beta,ftype] = kaiserord(fcuts,mags,devs,Fs);
n = n + rem(n,2);
hh = fir1(n,Wn,ftype,kaiser(n+1,beta),'scale');

figure(2)
freqz(hh,1, 4096, Fs)                                           % Filter Bode Plot

filt_sig = filtfilt(hh, 1, S);                                  % Filter Signal

FTS = fft(S)/L;                                                 % FFT Of Original Signal
FTFS = fft(filt_sig)/L;                                         % FFT Of Filtered Signal
Fv = linspace(0, 1, fix(L/2)+1)*Fn;                             % FFT Frequency Vector
Iv = 1:length(Fv);                                              % Index Vector

[pks,Frs] = findpeaks(abs(FTS(Iv))*2, Fv, 'MinPeakHeight',0.1); % Find Pure Tone Frequencies

figure(3)
semilogy(Fv, (abs(FTS(Iv)))*2)
hold on
plot(Fv, (abs(FTFS(Iv)))*2)
hold off
grid
axis([0  2500    ylim])

sound(filt_sig,Fs)

The Plots —

Bode Plot:

Frequency Plot Of Unfiltered & Filtered Signals:

[Vlad Dumitru]

Thanks, now it's working. But is nothing simple here xD.

[Star Strider]

My pleasure.

I doubt anything is simple anywhere, at least in maths and the sciences.

If my Answer solved your problem, please Accept it!

[VIVEKITHA]

could you plz explain how did you fix values for fcuts mags and devs

[Image Analyst]

Star - I accepted it for you.

[Star Strider]

@Image Analyst — Thank you!

@vv_moni — The ‘fcuts’ and ‘mags’ vector elements are dictated by the filter design (what I want the filter to do). The ‘devs’ elements are my arbitrary choices, depending on tightly I want the filter to match ‘mags’. The documentation for the kaiserord (link) function explains it better than I can.

[VIVEKITHA]

sorry for the late reply star strider Thank you!!I will check them

[Vincent Abraham]

Could you explain how u selected the values for Wp, Ws, Rp, Rs?

[Star Strider]

Vincent Abraham — This was a while ago, so since the signal was provided, I likely downloaded it and did a fft on it. On the basis of the fft results, I chose ‘Wp’, and from those, acceptable values for ‘Ws’. The ‘Rp’ and ‘Rs’ values can be anything reasonable. (These days, I’d use an elliptic filter for efficiency.)

Answer 2

[Image Analyst]

Did you try to apply the example in butter() to your code (like I did below):

% Read standard sample tune that ships with MATLAB.
[dataIn, Fs] = audioread('guitartune.wav');

% Filter the signal
fc = 800; % Make higher to hear higher frequencies.
% Design a Butterworth filter.
[b, a] = butter(6,fc/(Fs/2));
freqz(b,a)
% Apply the Butterworth filter.
filteredSignal = filter(b, a, dataIn);

% Play the sound.
player = audioplayer(filteredSignal, Fs);
play(player);

It works, and it seems easy enough. What happened when you tried it? Did you get an error or something?

[Vlad Dumitru]

When I'm tring your solution my sound it's almost the same, the single difference it's volum. If I'm using a high value of fs I will get high volum and if I'm using a low value will be low volum, but the sound it's exactly the same. That's the reasouns I've post and my sound, because I've already tried this kind of solutions how I said in my question and I did not obtain nothing....

[Image Analyst]

No. It sounds different. If you set fc = 300 I guarantee you should definitely notice a difference than if you set fc = 18000, at least for the guitartune sound. Does the code work for the guitartune sound when you change fc???

Maybe all your sounds are lower, I don't know. But there should be some frequency cutoff where your sound will sound different. If all your sounds are lower than 800 Hz and you set fc = 1000 and then 15000 then you won't hear any difference because you don't have any frequencies up that high. But if you do have freqs that high, you should hear a difference. Or if all your freqs are less than 1000 then you should hear a difference between an fc of 150 and an fc of 1000. Don't you?

[Vlad Dumitru]

If I'm using 300 or less... I can't hear noting, everything over 300-400 it's exactly the same sound on differite volum.... I've post my sound on that question, could you try with that? I really don't know what's wrong here. Here it's my sound on dropbox and here on google drive. So if you have and idea, please let me know.

[Image Analyst]

Star said he couldn't get it and when I try to play the sound from either of those two web sites you listed they say it's not a good sound file, so I suspect it's corrupt. Please post a good copy with the paperclip icon.

[Vlad Dumitru]

Sound file