EE342.01
Problem Set 9
DUE M 04/13/98
Consider the speech data stored in the file speech.dat
. The data corresponds to the spoken sentence "Should we chase those
young outlaw cowboys?" You can listen to the sentence if your version
of matlab supports the sound() function by typing sound(speech,
1/T) where speech is the vector of data and T is the sampling rate.
The microphone signal was sampled at 8kHz resulting
in 18,970 samples that are stored in the data file. In order to operate
on the whole signal you will need to use a professional version of matlab
due to the large size. If a professional version of matlab is unavailable,
take the first 6421 samples corresponding to the phrase "should we chase."
To illustrate the use of the DFT and the effect of aliasing perform the
following operations.
-
Plot the signal versus time letting matlab connect the points (i.e., don't
use stem). This keeps the plot from looking too messy.
-
Using the DFT, plot the magnitude and phase spectra of the sampled signal
versus frequency in Hertz. What frequencies are the most prevalent?
-
Reduce the sampling rate to 4000Hz by removing every other sample
in the signal. Plot the undersampled signal.
-
Using the DFT, plot the magnitude and phase spectra of the undersampled
signal versus frequency in Hertz.
-
Comment on the effects of undersampling. If possible listen to the
original signal as well as the undersampled signal to hear the effects.
-
Separate the first three words (should, we, chase) from the original signal
sampled at 8000Hz into separate signals and plot each word's magnitude
and phase spectra using the DFT. Based on this spectra data could
we do word recognition, i.e., distinguish which word was said by its spectra?