You have the sampling frequency as an output from the audioread function. The time vector would typically go
t = 0:1/fs:(length(soundfile)-1)/fs;
So then, 37 seconds will occur at roughly (37*fs + 1) in t. But this won't always work. Safer is to search for the closest value in your t variable.
[~,loc37] = min(abs(t-37));
[~,loc39] = min(abs(t-39));
x(loc37:loc39) = []; % this will cut out the sections of your sound file between 37 and 39 seconds
