Digitally Synthesizing Music in MATLAB

I’ve been following along in the textbook DSP First and saw a cool example of making digital music through MATLAB. It’s a nice bridge for me into DSP and eventually RF modulation. Here’s what I came up with.

Making Matlab Music!

Background

MATLAB has many different toolboxes available for Digital Signal Processing. I recommend installing Communications, Signal Processing, and DSP Systems. For this example, you only need the sound() function.

Also, I used a standard numerical keyboard to create my sounds in reference to the A4=440 Hz key. So a piano goes from 1-89, with middle C being key “40.” Here’s a picture of all the frequencies on a keyboard.

Middle C = key 40

Code Setup

The basic unit of this program is the sinusoid. The sine wave produces what we hear as music - it just uses an audible frequency (20-20,000 Hz). The basic wave equation is a function of just time with normalized inputs for our uses.

sinusoid = sin(2pi*frequency*t)

From here, you can mix, combine, and synthesize these signals. MATLAB treats them as a vector, meaning they can be added with vector dimensions. The function can also accept phase shifts or wave numbers if you want.

Code

As usual, you can find the code on github. It follows pretty directly from DSP First, first edition, page 436. The Synthesis_Signals is the main code, and Tone_Function is the function used to generate a code. There’s also a .mp3 file of just the melody to an example song. Here’s some important highlights:

First, here’s the ToneFunction. It takes the key name from a standardized list (keynum), and finds the frequency using this formula:

freq = 440 * 2^((keynum-49)/12);

Where 49 is the default tuning A=440Hz. That way each note is a magnitude of 2 above or below 440 Hz. The rest of ToneFunction is below.

function [tone] = ToneFunction(keynum,dur)
%NOTE Produce a sinusoidal waveform corresponding to a given piano key
%number
%   usage: tone = note (keynum, dur)
%   tone = the output sinusoidal waveform
%   keynum = the piano keyboard number of the desired note
%   dur = the duration (in seconds) of the output note
fs = 11025;
tt = 0:(1/fs):dur;
freq = 440 * 2^((keynum-49)/12);
tone = 1*sin(2*pi*freq*tt);
end

On the main file, I used a really cumbersome way to read music. I start out with two arrays: note and duration. I input all my notes in one, and all my duration (I used 60 bpm, so one quarter note is 1 second) in the other. This requires a huge array and attention to detail.

notes = [E5, Ds5, E5, Cs5, Ds5, E5, Ds5, Cs5, Ds5...
durrh = [.25, .25, 1.5, .25, .25, .5, .25, .25...

For the actual synthesis, the code uses a loop to add each tone into a sequential array.

for kk = 1:length(notes)
        keynum = notes(kk);
        dur = durrh(kk);
        tone = ToneFunction(keynum,dur);
        n2 = n1 + length(tone)-1;
        xx(n1:n2) = xx(n1:n2) + tone;
        n1=n2;
end

The code repeats itself for the amount of notes present in the array. It stacks all of this information into xx. For two simultaneous notes, I just added another note and duration array, and made the two end vectors have the same amount of items. Sound() produces this result with a default windows frequency of 11025 Hz.

for kk = 1:length(noteslh)
        keynum = noteslh(kk);
        dur = durlh(kk);
        tone = ToneFunction(keynum,dur);
        n2 = n1 + length(tone)-1;
        yy(n1:n2) = yy(n1:n2) + tone;
        n1=n2;
end
newy = zeros(size(xx)); %this creates a new matrix for RH and LH to make equal length (fills in extra zeros)
newy(1:size(yy,1),1:size(yy,2)) = yy;
newx = zeros(size(newy));
newx(1:size(xx,1),1:size(xx,2)) = xx;
sound (newx+newy, fs);

So it works with two notes. If you severed 8 of your fingers this program could be very applicable.

Conclusion and Application

Sergei Rachmaninoff is one of my favorite composers, so I gave him a tribute by synthesizing two lines to his “Vocalise.” His work is in the public domain so I’m not breaking any laws by posing my version here.

Obviously the code sounds a little generated. There are further expansions also listed in DSP First, like selective fading on each tone. I will try them out soon.

Overall I’d like to improve the synthesis and construction of 2+ notes. It seems pretty cumbersome, but I need more knowledge of vectors and DSP sinusoids to make that happen. If you have any suggestions, let me know!