Speaker Identification using machine learning

Hello,

Im writing a code on speaker identification systems and Im stuck at creating triangular mel filter bank. I got a wierd triangular filter bank when i plotted it. Please help any advice appreciated, thank in advance. The code and the Triangular filter plot are attached below.

% read voice file
[audio,fs] = audioread('B1.wav');


%resamplying fs to 16kHz 
fs1 = 16000;
[p,q] = rat(fs1/fs);
new_audio = resample(audio,p,q);
plot(new_audio)


%Pre-emphasis
a=[1];
preemph=[1 -0.95];
x = filter(preemph,a,new_audio);
plot(x)


%Frame blocking
fd = 0.025;
frameoverlap = 0.01;
N = length(x);
framelength = round(fd * fs1);
framestep = round(frameoverlap * fs1);
N_frames = floor(N/framelength);

%creating frames
frames=[];
for i=1:N_frames
frame = x((i-1)*framestep + 1: ((i-1)*framestep)+framelength);
max_val=max(frame);
   if (max_val>0.025)
frames=[frames;frame];
end
end

% Hamming window
plot(hamming(length(frames)))
win_frame = frames * hamming(length(frames));
plot(win_frame)


%FFT
ft = abs(fft(win_frame));
       plot(ft);


% Mel Filter Bank
fl = 0;
fh = fs1/2;
N_filter = 12;
nfft = 512;

%computing band in mel-scale
mel_low = 2595*log10(1+(fl/700));

mel_high=2595*log10(1+(fh/700));


%creating the mel-scaled vector
mel = linspace(mel_low,mel_high,N_filter+2);

% mel to Hertz warping function
melinhz=700*(exp(mel/2595)-1);

%Converting to frequency resolution
f = floor((nfft+1)*melinhz/fs1); 

%Creating triangular filter
for m =2:length(mel)-1
        for k=1:nfft
     if k<f(m-1)
        H(m-1,k) = 0;
    elseif (k>=f(m-1)&&k<=f(m))
        H(m-1,k)= (k-f(m-1))/(f(m)-f(m-1));
    elseif (k>=f(m)&&k<=f(m+1))
       H(m-1,k)= (f(m+1)-k)/(f(m+1)-f(m));
    elseif k>f(m+1)
        H(m-1,k) = 0;   
end
end
end

Answer 1

[Dat Le]

Could you give me your code about speaker identification?