Recreating Sine Wave from pitch detection

Dear Community, I am looking for a way to rebuild a signal from the pitch information from an other signal. Best scenario would be to read in a sine wave with changing pitch, extract the pitch information and then to be able to apply it to a sine wave. Any help is very much appreciated! Thanks!

Answer 1

[jibrahim]

Hi Hans,

Here is some code that approximates what you want. It uses:

• audioOscillator to create a frequency-varying sine wave signal
• pitch to estimate the pitch of a signal
• timescope to visualize the original and reconstructed signals.

% Create the original frequency-varying sine wave signal
src = audioOscillator;

F = [100:500 499:-1:100];
% We will change the frequency every 10 frames
numFramesPerFreq = 10;

frameLength = src.SamplesPerFrame;
sig = zeros(numFramesPerFreq*size(F,2)*frameLength,1);

% Build the sine wave
for index=1:size(F,2)
    src.Frequency = F(index);
    for index2=1:numFramesPerFreq
        startInd = (index-1)*numFramesPerFreq*frameLength+(index2-1)*frameLength + 1;
        sig(startInd:startInd+frameLength-1) = src();
    end
end

% Estimate the pitch of the signal
% A pitch estimate is generated for every "hop" of samples
p = pitch(sig,src.SampleRate,WindowLength=1024,OverlapLength=896);

% Hop length is window length - overlap length
hopLength = 1024-896;

% This is the reconstructed signal (based on pitch estimates)
sig2 = zeros(hopLength*size(p,1),1);

% We will build this signal
src = audioOscillator;
frameLength = hopLength;
src.SamplesPerFrame = hopLength;

for index=1:size(p,1)
    % For every frame, use estimated pitch
    osc.Frequency = p(index);
    sig2((index-1)*frameLength+1:index*frameLength) = src();
end

% Make original and reconstructed signals the same length
L = min(size(sig,1),size(sig2,1));
sig = sig(1:L);
sig2 = sig2(1:L);

% Use a scope to visualize the signals
scope = timescope(SampleRate=src.SampleRate,...
                  LayoutDimensions=[2 1],...
                  NumInputPorts=2,...
                  TimeSpan=.1,...
                  YLimits=[-1 1]);
scope.ActiveDisplay = 2;
scope.YLimits = [-1 1];

numFrames = size(sig,1)/hopLength;
for index=1:numFrames
    in1 = sig((index-1)*hopLength+1:index*hopLength);
    in2 = sig((index-1)*hopLength+1:index*hopLength);
    scope(in1,in2)
end

[Hans Buchele]

Hey Jibrahim,

Thanks for your suggestions but I am not getting consistent results. If I create an sinus wave of 220 hertz, the pitch from [f0,loc] = pitch(x,fs) is displayed correctly, but with higher frequencies, for example 1000 hertz, the results are around 330 hertz. It must be user error but I can not figure out the problem.

[jibrahim]

Can you include your code? That would help.

[Hans Buchele]

Sorry I wasn't on my computer before.

clc;

clearvars;

A = 1;

fs = 44100;

dt = 1/fs;

t = 0:dt:10;

f1 = 320;

x = A*sin(2*pi*f1*t);

y = A*sin(2*pi*f1*t);

filename = sprintf('%d.wav',f1);

stereo_mtx = [x(:), y(:)];

audiowrite(filename, stereo_mtx, fs, 'BitsPerSample', 24);

x = audioread(filename);

[f0,loc] = pitch(x,fs);

[jibrahim]

Hi Hans,

I can reproduce this. It seems you are hitting a limitation of the pitch function. Its algorithm works well for speech signals, but misfires for such test signals. I will gather more info on this soon.

In the meantime, you can use the AI-powered pitchnn function instead.It behaves much better with such signals.

You first download some required files:

downloadFolder = fullfile(tempdir,'crepeDownload');
loc = websave(downloadFolder,'https://ssd.mathworks.com/supportfiles/audio/crepe.zip');
crepeLocation = tempdir;
unzip(loc,crepeLocation)
addpath(fullfile(crepeLocation,'crepe'))

Then you should be able to use pitchnn:

clc;
clearvars;
A = 1;
fs = 44100;
dt = 1/fs;
t = 0:dt:10;
f1 = 1000;
x = A*sin(2*pi*f1*t);
y = A*sin(2*pi*f1*t); 
filename = sprintf('%d.wav',f1); 
stereo_mtx = [x(:), y(:)];
audiowrite(filename, stereo_mtx, fs, 'BitsPerSample', 24);  
x = audioread(filename);
pitchnn(x,fs); % No poutput specified gives you a plot
[f0,loc] = pitchnn(x,fs);

[jibrahim]

A little more info. For pure tones, you should use instfreq. All of the pitch algorithms are exploiting the harmonic structure of voice--they're really not designed for pure tones, and simpler algorithms like instfreq will perform best:

A = 1;
fs = 44100;
dt = 1/fs;
t = 0:dt:10;
f1 = 1000;
x = A*sin(2*pi*f1*t);
y = A*sin(2*pi*f1*t); 
filename = sprintf('%d.wav',f1); 
stereo_mtx = [x(:), y(:)];
instfreq(stereo_mtx,fs);

More info on the pitch function: It has a range parameter which is set by default to handle speech signal (default range is [50, 400] Hz). You can adjust the range for your case:

A = 1;
fs = 44100;
dt = 1/fs;
t = 0:dt:10;
f1 = 1000;
x = A*sin(2*pi*f1*t);
y = A*sin(2*pi*f1*t); 
filename = sprintf('%d.wav',f1); 
stereo_mtx = [x(:), y(:)];
pitch(stereo_mtx,fs,Range=[50 5000]);

The function pitchnn should work for both types of sound (speech and sine waves are fine)

[Hans Buchele]

Great thank you!

Answer 2

[Image Analyst]

Attached is a demo that creates a sound. Perhaps you can adapt it to change the frequency and amplitude envelope to whatever you want.