MFCC

function mel = freq2mel(freq)

mel = 2595*log10(1+(freq/700));

end

function freq = mel2freq(mel)

freq = 700*(10^(mel/2595)-1);

end

NOTE

MEL spectrogram is better suited to represent the human perception of the sound and can be useful in speech recognition and speech processing tasks.

Spectrogram is more widely used in signal processing and audio analysis.

Activity 4.1

Create Mel Filter Banks

f_min = 1000;

f_max = 8000;

n_bands = 6;

mel_min = round(freq2mel(f_min))

mel_max = round(freq2mel(f_max))

bound_range = linspace(mel_min,mel_max,n_bands+1)

center_mel_freqs = zeros(1,n_bands)

for i=1:length(bound_range)-1

center_mel_freqs(i) = round((bound_range(i)+bound_range(i+1))/2);

end

disp(center_mel_freqs)

i = 2

x = [bound_range(i),center_mel_freqs(i),bound_range(i+1)]

y = [0,1,0]

plot(x,y)

axis([bound_range(1) bound_range(end) 0 1])

figure()

for i=1:n_bands

x = [bound_range(i),center_mel_freqs(i),bound_range(i+1)]

y = [0,1,0]

plot(x,y,'LineWidth',3)

hold on

axis([bound_range(1) bound_range(end) 0 1])

xticks(round(bound_range))

end

xlabel('Mel Frequencies')

title('Mel Filter Bands')

Reference

https://www.youtube.com/watch?v=4_SH2nfbQZ8&ab_channel=ValerioVelardo-TheSoundofAI