Description
1. Load a speech signal of your interest and select voiced region then Find pitch using LP
analysis? Plot the results (Speech, LP-residual, autocorrelation of LP-residual) Note:
Computer Based Question [5 pts]
2. For this Question you will use the code in the Matlab file: lpc try.m. Record a speech
signal using your voice in wav format. You may use fs = 16000 as sampling frequency
during your recording. Note: Computer Based Question (MATLAB) [15 pts]
More specifically:
(a) Analysis-Synthesis based on Linear Prediction
Download the Matlab code lpc try.m
At this stage, the lines require your interventions are 44-47.
• In 44, you are require to compute the autocorrelation function of a given speech
frame defined in line 40. You may use the Matlab function xcorr.
• In line 45 you must compute the autocorrelation coefficients using the Levinson
recursion. In Matlab, there is a function lpc which can be used.
• In line 46 you must compute the gain of the LP filter, as discussed in lectures.
• In 47, you must compute the linear prediction error, by inverse filtering the
speech signal through the estimated linear prediction filter. You may use the
Matlab function filter.
Once done, you must be able to load the wav file, perform analysis and synthesis
frame-by-frame. At the end you can save your computed speech signal (variable
out which is the output from the above Matlab file. To save it, you may use the
command audiowrite). Listen to the original and the processed speech signal using
the Matlab command soundsc or sound.
Listen to output audio file and briefly write down your observations.
(b) Follow frame-by-frame the analysis procedure
In each analysis frame, compare the magnitude of the frequency response of
the estimated linear prediction filter (using Matlab command freqz) with the
magnitude of the Fourier Transform (Matlab command fft) of the speech frame.
• Plot and compare on the same figure (use hold on and hold off – use different
colors) the two magnitude spectra. Do the above in both voiced and unvoiced
areas.
• Change the order of the Linear Prediction towards both directions (by increasing
and decreasing the initial value (line 24: 24(!)))
Save some interesting – according to you – plots and write down briefly
your comments/observations.
Note: This assignment is strictly based on MATLAB
