You are on page 1of 3

738015

Digital Signal Processing Applications


Postgraduate Course
Auckland University of Technology

(2010)

Assignment 1

This assignment includes problems and MATLAB exercises from Chapter 2 of your text
book and lab manual as well as a project on speech processing.
Note: You can download the Lab Manual from AUT online, Course Notes.

Due date: Monday, 9 August 2010, at 2 pm.


Mark: 15% of the total mark.

A.1 Problems: [3 marks]


Solve problems 2.30, 2.41, 2.44, 2.61, 2.72, and 2.90 from Chapter 2 of your textbook.

A.2 MATLAB exercises: [4 marks]


Perform MATLAB exercises M2.1, M2.3, M2.5, M2.7, and M2.9 from Chapter 2 of the
text book.
Perform MATLAB exercises Q1.27, Q1.28, Q1.29, Q1.32, Q1.33, Q2.19, Q2.20, Q2.29,
Q2.34 from Chapters 1 and 2 of the lab manual.
Note: Include comments and explanation about the main function of each Matlab
code and submit mfiles as well as figures related to each exercise.

A.3 Speech Processing Project: [8 marks]

Reverberation (Reverb):
Reverb can basically be defined as the effect of reflected sound that occurs in a room. As
a result Digital Signal Processors are often employed to simulate the effects of rooms
(areas) which cannot be recreated easily for applications such as the musical effect of a
concert or large hall.

When we hear sound in a hall (for example), the first sound the ear will receive will be
the direct sound from the speaker, following the initial direct sound, some time later a
slightly delayed and attenuated version will be heard from reflection off the near walls.
Further reflections will create further attenuated and delayed versions of the initial sound.

The diagram below gives a rough approximation of the impulse response of a room,
where the first sound heard is the direct sound then the subsequent impulses are the

1
reflected and attenuated versions of the same sound where the room will eventually
absorb the original sound created.

Figure 1

To create this effect we can use a Feedback comb filter shown in figure 2 below. The
impulse response of a comb filter has the response of resembling the reflection of a sound
wave around a room with decreasing amplitude. A single echo can be created using a
feed forward comb filter.

Figure 2

Y(n) = X(n) + gY(n – N)

To maintain stability the feedback gain must be less than 1.

2
A single comb filter is not enough to create a natural sounding effect of reverb, so it is
common to have four or more comb filters in parallel with varying delays and gains to
create the complex reflections found in a typical room.

Figure 3

The two allpass filters in series increase the density of the reflected frequencies. Figure 3
shown above is just one of many methods of implementing a reverb filter, but they share
the common building blocks of the comb and allpass filters.

Project procedure:

i. Create a speech file by recording your voice for about 30 seconds (Alternatively
you may use one of the sample files from AUT online, DSPA, Resources) and
save it as “myfile.wav”.
ii. Use Matlab to generate a noise signal of the same size as your speech file and add
this noise to the speech signal to create the noisy version of your speech file.
iii. Save the noisy speech file as “mynoisy.wav” for further processing.
iv. Use Matlab to play both the original and the noisy speech files.
v. Plot the first five seconds of both the original and the noisy signals in time-
domain and compare the plots.
vi. Repeat the previous step (step v) for the frequency-domain representation. In
another word, plot the power spectrum of the first five minutes of the two signals.
vii. Plot the spectrogram of the noisy signal and compare it with the spectrogram of
the original signal. Comment on the differences.
viii. Design a proper filter to remove noise from the noisy speech signal and compare
the noise removed signal with the original one. You can plot the difference of
these two signals and show that the difference is small.
ix. Apply downsampling of 2 to the noise removed signal and state the effect of this
operation on the signal by listening as well as plotting it in time domain.
x. Apply upsampling of 2 to the nose removed signal and state the effect of this
operation on the signal by listening as well as plotting it in time domain.
xi. Apply a proper reverberation method to the noise removed signal to add echo to
the signal and save the result in a file called “Myecho.wav”.
xii. Discuss other possible sound effects that can be added to this project.

*****************************

You might also like