returned T in spectrogram()

Hi All!

Suppose that I want to divide a signal in segments and I want to calculate the center of those segments and the respective time. Suppose that I' ve

signal = [1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17];
fs_audio = 0.5; % sampling frequency = 0.5 HZ

winLen_Sec_desired = 5 % length of a segment in seconds = 5 seconds
winOverlap_desired = 0.4 % percentage of overlapping segments = 40%

depending on the sampling frequency the desired parameter won't be the real ones, but it's ok...

winLenSamp = round(winLenSec_desired*fs_audio); % length of a segment in
                                                % samples
if mod(winLenSamp,2) == 0   % check that the segment has an odd size
    winLenSamp = winLenSamp+1;    
end
winLenSec_Real = winLenSamp/fs_audio; % our window size will be slightly
                                      % different from the one desired

winOverlapSamp = ceil(winLenSamp*winOverlap_desired);
winOverlap_Perc = winOverlapSamp/winLenSamp; % our overlap will be slightly
                                             % different from the one desired

winStep = winLenSamp - winOverlapSamp; % we'll have a center each winStep
                                       % samples

firstCenter = ceil(winLenSamp*0.5);    % calculate the first center

j = 1;                                 
i = firstCenter;                       
while (i + winLenSamp - firstCenter) <= length(signal) %till we exceed 
                                                       % the dimension
    winCenters(j) = i;     %save in winCenters the center of the segment j
    j= j +1;                           % increment the segment
    i= i + winStep;                    % calculate the new center position
end

time = (0:length(signal)-1)/fs_audio;
winTime = time(winCenters);

I hope that till here everything is correct... Anyway I'm interested in winTime and winCenters

Now I have to use the spectrogram function in matlab to compute the STFT of the signal. In particular I use

VECTORFREQ = [150:10:400];
[S F T P] = spectrogram(signal, rectwin(winLenSamp), winOverlapSamp, VECTORFREQ, fs_audio, 'yaxis');

in order to have back the vector T containing the time instants for each segment in which the FFT is calculated.

I guess that T and winTime calculated before should be the same but they are not! Any guess?

Thanks a lot Gabriele

ps I know that to divide a signal in segments I can use the function buffer, but it doesn't return the center of the segments nor the time where each segment is centered.

Answer 1

[Youssef Khmou]

Hi Gabriel :

I tested your code , it gives :

 >> T

 T =

     3     5     7     9    11    13    15    17    19    21    23    25    27      29    31

>> winTime

 winTime =

     2     4     6     8    10    12    14    16    18    20    22    24    26    28    30

I guess that the winLen_Sec_desired equals 5, the real one winLenSec_Real equals 6 so that shift is seen between T and winTime

Temporal solution :

Increment with 1 this line in your code :

 winTime = 1+time(winCenters);

they, now, give same result .

[Gabriele]

Thanks for the answer!

I don't know if this is the reason since I'm giving as input of spectrogram the winLenSamp (windown size in samples not time) and the winOvelapSamp (overlap in samples). They are the same values that I used to calculate the winTime in the for loop...so I was expecting the same results....

[Gabriele]

I've tested also with other data and the shift is not

winLenSec_Real-winLenSec_desired;

for all the values in T

[Youssef Khmou]

hi,

Post the counter example to see ,

with incremented winTime, it gave same vectors with these elementary signals :

 %signal = [1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17];
%signal=sin(2*pi*(0:2:100)*0.25);  % Fc=0.25 Hz
%signal=rand(10,1)';
%signal=square(2*pi*(0:2:100)*0.13); Fc=0.13 Hz
%signal=exp(-(0:2:10)).*cos(2*pi*0.10*(0:2:10)); % Fc=0.1 Hz
%signal=exp(j*2*pi*0.33*(0:0.1:10)); % Fc=0.33 Hz 
signal=randint(10,1);