Your Question leaves some details undefined.
However:
You choose the sampling frequency of your speech signal (ideally 44.1 kHz). See the documentation for audiorecorder for details.
The sampling interval is the inverse of that: Ts = 1/Fs;
The Nyquist frequency (usually Fn) is one-half the sampling frequency, Fn = Fs/2;. For reasons I won’t go into here, it is the highest uniquely resolvable frequency of a sampled signal.
Bandwidth of a speech signal varies, but for engineering purposes is considered to be 6 kHz.
The R2015a documentation for fft clearly explains what you need to do to calculate and plot it. Note specifically the code between the top two plot figures.
Of note, the 44.1 kHz sampling frequency is used because the upper limit of normal human hearing is about 20 kHz. The relevant Nyquist frequency slightly exceeds that.
