0.2 Another way of viewing the world  (Page 2/2)

From this type of observation, a whole slew of theory and algorithms have been developed to characterize such behavior. By taking the Fourier Transform of our sound file, we can see what frequencies make up the signal. Another way of thinking about it, the sound file can be decomposed as the sum of waves of differing frequencies, and the Fourier Transform provides a way of seeing the sound file in the time domain, as in Figures 2 and 3, to the frequency domain shown below in Figure 4.

The spectrogram shows us how the signal can be broken down into an infinite sum of waves with different frequencies. As Figure 4 shows, the most dominant frequency occurs at about 220Hz. This corresponds to the second harmonic of the signal. The fundamental frequency being 110Hz, which is correct, since the 'A' string of the guitar should be tuned to 110Hz. Apparently the guitar was in tune when the recording was made. Furthermore, you will also notice other dominant waves with frequencies of about 330Hz and 440Hz. Of the musicians out there reading this, this makes sense since 330Hz corresponds to an 'E' which would be the fifth of the 'A' and the third harmonic to the 'A' at 110Hz. But the take away message from these figures is that there are two ways of seeing the same phenomenon of the recorded sound file. One is to simply plot the resulting signal versus time. Another is to view the same signal and its spectrogram, or rather, what frequency components make up the sound file.

Plucking the 'e' string of a guitar

Similar to plucking the guitar's 'A' string, here we provide the same graphs but for the sound file in Figure 5.

Let's point out some differences between Figures 2 and 6. Notice that in Figure 2, the signal decays to 0 a couple of seconds after the signal in Figure 6. Also notice that in Figure 7, the spacing between the oscillations of the signal is smaller than those in Figure 3.

Also, again notice that in Figure 7, the zoomed-in graph of the signal versus time, there are recurring patterns in the signal. See if you can convince yourself that the signal can be described as the sum of waves with differing frequencies.

Figure 8 shows the spectrogram of the sound file in Figure 5. The dominant peak occurs at 330Hz, which from our previous discussion is not surprising. The high 'E' string of a guitar should be tuned to 330Hz.

However, there are some differences to note between Figures 4 and 8. The spectrogram in Figure 8 has less prominent peaks. Between 200 and 300Hz, there seems to be a lump of frequencies that our sound wave has. One of the reasons the spectrum is not as "clean" as the one in Figure 4, is because the high E string causes the lower strings on the guitar to vibrate. Thus one sees lumps in the spectrogram below the fundamental frequency of 330Hz as well as the harmonics of the frequencies of the lower strings.

Extra readings

For more information on the algorithm/procedure that takes you to/from the frequency domain, Derivation of the Fourier Transform and The Fast Fourier Transform (FFT) are good mathematical sources.