Computing Audio Spectra in Matlab

• The fft function computes the FFT of a specified signal.

• In general, we will want to view either the magnitude or phase values of the FFT coefficients, which in Matlab can be determined using the abs and angle functions applied to the FFT'ed data.

• The spectrum of a single-channel soundfile can be plotted as:

  % This assumes the soundfile is a single channel
  [y, fs] = audioread('soundfilename');
  Y = fft(y);
  N = length(Y);
  f = fs*(0:N-1)/N;
  %plot(f, abs(Y))           % plot linear magnitude
  %plot(f, angle(Y))         % plot phase
  plot(f, 20*log10(abs(Y))); % plot magnitude in dB
  xlim([0 fs/2])
  xlabel('Frequency (Hz)');
  ylabel('Magnitude (dB)');
  

• A variety of windows can be applied to a signal before the computation of the FFT using the functions hann, hamming, blackman. For a complete list, see the window function help. Time-domain windows, which will be discussed in more detail later in the course, can help minimize spectral artifacts related to signal truncation.

• The spectrogram function computes a time-frequency plot of a signal where color represents spectral magnitude amplitude.