Extract Voices from Music Signal

Implement a basic digital music synthesizer and use it to play a traditional song in a three-voice arrangement. Specify a sample rate of 2 kHz. Save the song as a MATLAB® timetable.

fs = 2e3;
t = 0:1/fs:0.3-1/fs;

l = [0 130.81 146.83 164.81 174.61 196.00 220 246.94];
m = [0 261.63 293.66 329.63 349.23 392.00 440 493.88];
h = [0 523.25 587.33 659.25 698.46 783.99 880 987.77];
note = @(f,g) [1 1 1]*sin(2*pi*[l(g) m(g) h(f)]'.*t);

mel = [3 2 1 2 3 3 3 0 2 2 2 0 3 5 5 0 3 2 1 2 3 3 3 3 2 2 3 2 1]+1;
acc = [3 0 5 0 3 0 3 3 2 0 2 2 3 0 5 5 3 0 5 0 3 3 3 0 2 2 3 0 1]+1;

song = [];
for kj = 1:length(mel)
    song = [song note(mel(kj),acc(kj)) zeros(1,0.01*fs)];
end
song = song'/(max(abs(song))+0.1);

% To hear, type sound(song,fs)

tune = timetable(seconds((0:length(song)-1)'/fs),song);

Open Signal Analyzer and drag the timetable from the Workspace Browser to the Signal table. Click Display Grid to create a two-by-two grid of displays. Add a spectrum view to the top two displays and the lower left display. Select the lower right display, click Time-Frequency to add a spectrogram view, and click Time to remove the time view. Drag the song to all four displays. Select the lower right display, and on the Spectrogram tab, specify a time resolution of 0.31 second (310 ms) and 0% overlap between adjoining segments. Set the Power Limits to –50 dB and –10 dB.

On the Analyzer tab, click Duplicate three times to create three copies of the song. Rename the copies as high, medium, and low by double-clicking the Name column in the Signal table. Move the copies to the top two and lower left displays.

Preprocess the duplicate signals using filters.

1. Select the high signal by clicking its name in the Signal table. On the Analyzer tab, select Highpass from the Preprocessing gallery. On the Highpass tab that appears, enter a passband frequency of 450 Hz and increase the steepness to 0.95. Click Highpass.

2. Select the medium signal by clicking its name in the Signal table. On the Analyzer tab, select Bandpass from the Preprocessing gallery. On the Bandpass tab that appears, enter 230 Hz and 450 Hz as the lower and upper passband frequencies, respectively. Increase the steepness to 0.95. Click Bandpass.

3. Select the low signal by clicking its name in the Signal table. On the Analyzer tab, select Lowpass from the Preprocessing gallery. On the Lowpass tab that appears, enter a passband frequency of 230 Hz and increase the steepness to 0.95. Click Lowpass.

View a spectrogram on each of the three displays containing filtered signals.

1. Remove the original signal by clearing the check box next to its name.

2. On the Display tab, click Time-Frequency to add a spectrogram view and click Time to remove the time view.

3. On the Spectrogram tab, specify a time resolution of 0.31 second and 0% overlap between adjoining segments. Set the Power Limits to –50 dB and –10 dB.

Select the three filtered signals by clicking their Name column in the Signal table. On the Analyzer tab, click Export and save the signals to a MAT-file called music.mat. In MATLAB, load the file to the Workspace. Plot the spectra of the three signals.

load music

pspectrum(low)
hold on
pspectrum(medium)
pspectrum(high)
hold off

% To hear the different voices, type 
% sound(low.low,fs), pause(5), sound(medium.medium,fs), pause(5), sound(high.high,fs)