MATLAB Functionality for Digital

Speech Processing
• MATLAB Speech Processing Code
• MATLAB GUI Implementations

Lecture_3_2013 1
Graphical User Interface

GUI Lite 2.6

2
 Basics
Waveform Strips Plot

How do we rapidly and efficiently create a GUI for problems like

the one shown above? 3
Graphical User Interface Components
• GUI Lite created by students at Rutgers University to simplify the process
of creating viable GUIs for a wide range of speech and image processing
exercises
• GUI Lite Elements
– basic design tool and editor (GUI Lite 2.5/2.6)
– panels; used to block group of buttons/graphical panels/etc., into one or more
coherent blocks
– graphics panels; used to display one or more graphical outputs (figures)
– text block; used to display global information about the specific speech
processing exercise
– buttons; used to get and set (vary) exercise parameters; used to display a list
of exercise options; used to initiate actions within the code
• editable buttons – get and/or set parameter value
• text buttons – display variable values
• slider buttons – display variable range
• popupmenu buttons – display list of variable options (e.g., list of speech files)
• pushbuttons – initiate actions within the code
4
 Basics
Zoom Waveform Strips Plot
Text Box 1

Graphics
Panel 1
Buttons 1-11

Panel 2 Graphics
Panel 2

Panel 1 Panel 3 5
 GUI LITE 2.6 Setup Process
• Create a directory for loading all necessary file folders for creating GUI 26
exercises; i.e., call this directory:
‘matlab_central_speech’
• Define full path to the chosen directory; i.e.,
path_to_speech=‘C:\data\matlab_central_speech’
• Download (from MATLAB Central), and place in the chosen directory, the
following code and data folders:
– functions_lrr; speech_files; highpass_filter_signal; VQ;
– cepstral coefficients; isolated_digit_files; GUI_LITE_26;
• Download (from MATLAB Central) the folder:
‘pathnew_matlab_central.m’
• Run pathnew script to set up path links to GUI LITE 26 and to the
directories above
6
 GUI LITE 2.6 Setup Process
• Search for MATLAB Central using web search tool with
search input ‘MATLAB Central’ and go to MATLAB
Central
• Click on File Exchange
• In the File Exchange website type ‘speech processing
exercises’ in the search box
• Find and download each of the folders from the
previous vu-graph;
– find the folder of interest
– click on download
– from the displayed zip file, choose extract and search for
the matlab directory and download all files

7
 GUI LITE 2.6 Design Process
• begin with a rough sketch of the GUI 2.6 output, segmented
into button panels, graphics panels, text boxes, and buttons
• create exercise folder; e.g., ‘hello_goodbye_world’
• run program ‘runGUI.m’ to create GUI elements and save as a
GUI file (e.g., filename.mat);
• the software also creates an apps program named
filename_GUI26.m and a Callbacks program (just the GUI
structure) named Callbacks_filename_GUI26.m
• edit the program Callbacks_filename_GUI26.m by adding the
appropriate callback code for each of the graphics panels, text
boxes and button boxes
• run the resulting exercise and loop on GUI design and
Callbacks implementation
8
Hello/Goodbye World Plan

Design Specs:
• 2 Panels (for linking
inputs and outputs)

• 1 Text Box (for

describing the Exercise
GUI)

• 3 Buttons (all
pushbuttons) (for
embedding Callback
code to play two
messages and to close
up the GUI)

9
GUI26 Initial Screen

10
 GUI26 Creation for ‘hello_goodbye_world’
• run the program ‘runGUI.m’ and click on the ‘New’ button
• enter values for the number of panels (2), number of graphics
panels (0), number of text boxes (1), and number of buttons (3)
• enter the name for the GUI .mat file (‘hello_goodbye_world’)
• automatically create the set of GUI objects specified above,
using the mouse cursor to define the range and properties of
each object (this step is guided by the program ‘runGUI.m’
• the resulting specifications for the GUI are automatically saved in
the designated .mat file, ‘hello_goodbye_world.mat’
• edit the Callbacks routine,
‘Callbacks_hello_goodbye_world_GUI26.m’, by entering the
MATLAB code for the graphics panels, text boxes and button
boxes 11
 GUI26 Callback Code – Button 1
% Callback for button 1 – print out message Hello World
function button1Callback(h,eventdata);

% title box
stitle1=strcat(‘Hello World Using GUI2.5’);
set(titleBox1, ‘String’, stitle1);
set(titleBox1, ‘FontSize’, 25);

% uiwait(msgbox(‘Hello World!’, ’Message1’, ‘modal’));

waitfor(errordlg(‘Hello World’,’Message1’));
return;
end
12
GUI26 Callback Code – Button 2
% Callback for button 2 – print out message Hello World
function button1Callback(h,eventdata);

% title box
stitle1=strcat(‘Goodbye World Using GUI2.5’);
set(titleBox1, ‘String’, stitle1);
set(titleBox1, ‘FontSize’, 25);

% uiwait(msgbox(‘Goodbye World!’, ’Message1’, ‘modal’));

waitfor(errordlg(‘Hello World’,’Message1’));
return;
end
13
GUI26 Callback Code – Button 3
% Callback for button3 -- Close GUI
function button3Callback(h,eventdata);

% title box
stitle1=strcat('Close GUI');
set(titleBox1,'String',stitle1);
set(titleBox1,'FontSize',25);

uiwait(msgbox('Closing GUI','Message2','modal'));
% waitfor(errordlg('Closing GUI','Message2'));
% return;

display Goodbye
close(gcf);
end 14
hello_goodbye_world

run:
hello_goodbye_world_GUI26.m

directory:
hello_goodbye_world

15
 hello_goodbye_world
Text Box1

Button1 - Pushbutton
Message Box from Code

Button2 - Pushbutton

Button3 - Pushbutton

Panel2 Panel1
16
 hello_goodbye_world – GUI
Modifications
• Run program ‘runGUI.m’ to bring up GUI Lite 2.6 editor
• Choose Mod (modify) and select GUI file
‘hello_goodbye_world.mat’ for editing
• Choose ‘Move & Resize Feature’ option
• Choose ‘Button’ option
• Left click inside button to be modified
• Choose new button coordinates by using graphics cursor to
identify lower left and upper right corners of modified
button
• Click ‘Save GUI’ button
• Iterate on other buttons
• Click ‘Quit’ option to terminate GUI Lite 2.6 editor
17
GUI Lite 26 Edit Screen

2 Panels
0 Graphics Panels
1 Text Box
3 Buttons
18
GUI LITE 2.6 Edit Screen

Add Feature

Delete Feature

Move & Resize Feature

runGUI.m
Modify Feature

Feature Index

Save GUI

Save GUI As

Quit

hello_goodbye_world_GUI26.m19
 GUI Lite 2.6 Features
• separates GUI design from Callbacks for each
GUI element
• provides a versatile editor for modifying GUI
elements without impacting the Callback
actions
• provides a GUI element indexing feature that
enables the user to identify GUI elements with
the appropriate Callback elements

20
 Missing GUIDE Features

• radio button • are the missing features of

• check box value?
• listbox • do we need these features?
• toggle button • can we create the desired
set of speech processing
• table
exercises without these
• axes features?
• button group • can we add these features
• active X control to the GUI LITE editor?

21
 Pyramid of Learning
Ability to implement
theory and concepts in
working code (MATLAB, C,
C++); algorithms,
applications
Basic understanding of
how theory is applied;
autocorrelation,
waveform coding, …

Mathematics, derivations,
signal processing; e.g.,
STFT, cepstrum, LPC, …

Need to understand speech processing at all three

levels 22
 The Speech Stack
Speech Applications — coding, synthesis,
recognition, understanding, verification, language
translation, speed-up/slow-down

Speech Algorithms— speech-silence (background),

voiced-unvoiced, pitch detection, formant
estimation

Speech Representations — temporal, spectral,

homomorphic, Linear Prediction Coding

Fundamentals — acoustics, linguistics, pragmatics,

speech production/perception
Basics – read/write speech/audio files; display speech files; play files 23
MATLAB Exercise Categories
• Basic MATLAB Functions for
handling speech and audio files

• Advanced MATLAB Functions for

Speech Processing

24
 MATLAB Exercise Categories
• The speech processing exercises are grouped
into 5 areas, namely:
– Basics of speech processing using MATLAB (5)
– Fundamentals of speech processing (6)
– Representations of speech in time, frequency,
cepstrum and linear prediction domains (22)
– Algorithms for speech processing (7)
– Applications of speech processing (17)

25
 Basic Functionality
• read a speech file (i.e., open a .wav speech file and read the speech sample into a
MATLAB array)
• write a speech file (i.e., write a MATLAB array of speech samples into a .wav
speech file)
• play a MATLAB array of speech samples as an audio file
• * play a sequence of MATLAB arrays of speech samples as a sequence of audio
files
• record a speech file into a MATLAB array
• plot a speech file (MATLAB array) as a waveform using a strips plot format
• * plot a speech file (MATLAB array) as one or more 4-line plot(s)
• convert the sampling rate associated with a speech file (MATLAB array) to a
different (lower/higher) sampling rate
• lowpass/highpass/bandpass filter a speech file (MATLAB array) to eliminate DC
offset, hum and low/high frequency noise
• plot a frame of speech and its associated spectral log magnitude
• plot a spectrogram of a speech file (MATLAB array)
• * plot multiple spectrograms of one or more speech files (MATLAB arrays)

* indicates exercise not yet done 26

 Read a Speech File into a MATLAB Array
• [xin, fs, nbits] = wavread(filename);
• [xin, fs] = loadwav(filename);
– filename is ascii text for a .wav-encoded file which contains a speech
signal encoded using a 16-bit integer format
– xin is the MATLAB array in which the speech samples are stored (in
double precision format)
– fs is the sampling rate of the input speech signal
– nbits is the number of bits in which each speech sample is encoded
(16 in most cases)
– program wavread scales the speech array, xin, to range −1≤xin≤1,
whereas loadwav preserves sample values of the speech file and
hence array xin is scaled to range −32768≤xin≤32767
• [xin1, fs, nbits] = wavread(‘s5.wav’);
• [xin2, fs] = loadwav(‘s5.wav’);
27
 Read a Speech File into a MATLAB Array
• % test_wavread.m
• % test waveread function
• %
• % read speech samples from file 'test_16k.wav' into array x1 using wavread
• % routine
• filein='test_16k.wav';
• [x1,fs1,nbits]=wavread(filein);
•
• % print out values of fs1, nbits, wavmin1, wavmax1
• wavmin1=min(x1);
• wavmax1=max(x1);
• fprintf('file: %s, wavmin/wavmax: %6.2f %6.2f, fs1: %d, nbits: %d \n‘,…
• filein,wavmin1,wavmax1,fs1,nbits);
•
• % read speech samples from same file into array x2 using loadwav routine
• [x2,fs2]=loadwav(filein);
•
• % print out values of fs2, nbits, wavmin2, wavmax2
• wavmin2=min(x2);
• wavmax2=max(x2);
• fprintf('file: %s, wavmin/wavmax: %d %d, fs2: %d \n',...
• filein,wavmin2,wavmax2,fs2);
Terminal Display:
file: test_16k.wav, wavmin/wavmax: -1.00 1.00, fs1: 16000, nbits: 16
28
file: test_16k.wav, wavmin/wavmax: -32768 32767, fs2: 16000
 Write a Speech Array into a Speech File
• wavwrite(xout, fs, nbits, filename);
• savewav(xout, filename, fs);
– xout is the MATLAB array in which the speech samples are stored
– fs is the sampling rate of the output speech signal
– nbits is the number of bits in which each speech sample is encoded
– filename is the ascii text for the .wav-encoded file in which the
MATLAB signal array is to be stored
– for wavwrite the MATLAB array xout needs to be scaled to the range
−1≤xin≤1 whereas for savewav the MATLAB array xout needs to be
scaled to the range −32768≤xout≤32767
• wavwrite(xin1, fs, ‘s5out.1.wav’);
• savewav(xin2, ‘s5out.2.wav’, fs);

29
 Play a Speech File
• sound(x, fs);
• soundsc(x, fs);

– for sound the speech array, x, must be scaled to the range −1≤x≤1
– for soundsc any scaling of the speech array can be used
– fs is the sampling rate f the speech signal

• [xin, fs] = loadwav(‘s5.wav’); % load speech from s5.wav;

• xinn = xin/abs(max(xin)); % normalize to range of − 1 to 1;

• sound(xinn, fs); % play out normalized speech file;

• soundsc(xin, fs); % play out unnormalized speech file;

31
 Play Multiple Speech Files

• *play_multiple_files.m;
– sequence of filenames read in via filelist, keyboard or file search

• Example of usage to play out 3 speech files in sequence:

– kbe=filename entry via filelist(2), keyboard(1), or file search(0):1; %
keyboard chosen
– N=number of files to be played in a group:3; % play out 3 files
– i=1; filename: s1.wav;
– i=2; filename: s2.wav;
– i=3; filename: s3.wav

32
 Play Multiple Speech Files
• test_play_files.m
– play the following sequence of files:

s2.wav
s3.wav
s4.wav
s5.wav
s6.wav
play_multiple_files_GUI26.m

33
Record Speech into MATLAB Array
• record_speech.m (calls MATLAB function
audiorecorder.m, formally wavrecord.m)

• function y=record_speech(fs, nsec);

– fs: sampling frequency
– nsec: number of seconds of recording
– y: speech samples array normalized to peak of
32767

34
Display Speech Waveform

Strips Plot
4-Line Plots

35
 Basics
Zoom Waveform Strips Plot
• Plotting and examining speech/audio waveforms is one
of the most useful ways of understanding the
properties of speech and audio signals.
• This MATLAB Exercise displays a speech/audio
waveform as a single running plot of samples (called a
Strips Plot).
• Exercise plots from designated starting sample to
designated ending sample, with a user-specified
number of samples/line.
• Zoom feature to select region of signal for display.
• Plots use either samples or seconds, as specified by the
user.

36
 Basics
Waveform Strips Plot

37
 Basics
Waveform Strips Plot – Zoom 1

38
 Basics
Waveform Strips Plot – Zoom 2

zoom_strips_plot_GUI25.m 39
*Plot Speech Using 4-Line Plot

40
 Sampling Rate Conversion
• y = srconv(x, fsin, fsout);
– x: input speech array;
– fsin: input speech sampling rate;
– fsout: desired speech sampling rate;

• Example:
– [xin, fsin] = loadwav(‘s5.wav’); % fsin=8000;
– fsout = 10000; % desired sampling rate;
– y = srconv(xin, fsin, fsout);

41
Sampling Rate Conversion Basics

SRC_GUI25.m 42
 Basics
Frame-Based Spectrums

NB_WB_spectra_GUI25.m 44
 Basics
Wideband/Narrowband Spectrogram

Spectrogram_GUI25.m 45
*Plot Multiple Spectrograms

46
 Fundamentals

• 2-tube vocal tract model

• 3-tube vocal tract model
• p-tube vocal tract model
• glottal pulse model and spectrum
• composite vocal tract model and spectrum
• ideal vocal tract model and spectrum

47
 Representations
• time domain exercises
– windows; features; autocorrelation estimates; amdf
• frequency domain exercises
– phase/magnitude; overlap-add windows; WSOLA
• cepstral domain exercises
– analytical cepstrum; single pole cepstrum; FIR
sequence cepstrums; cepstrum aliasing; cepstrum
liftering; cepstral waterfall
• linear prediction exercises
– LPC frames; LPC error; LPC varying p; LPC varying L;
LSP roots; plot roots

48
 Algorithms
• endpoint detector
• Voiced-Unvoiced-Background estimation
method
• autocorrelation pitch detector
• log harmonic spectral waterfall plots
• cepstral pitch detector
• SIFT pitch detector
• formant estimation method

49
 Applications – Part 1
• Speech waveform coding;
– statistical properties of speech; quantization
characteristics of a B-bit uniform or mu-law compressed
and quantized speech file; uniform quantization; mu-law
compression; mu-law quantization; Signal-to-Noise Ratio
(SNR) of uniform and mu-law quantizers
• Automatic Gain Control (AGC)
• Adaptive Differential Pulse Code Modulation (ADPCM)
waveform speech coder
• Vector Quantizer (VQ); VQ Cells
• Synthetic vowel synthesizer

50
 Applications – Part 2

• LPC error synthesis

• LPC vocoder
• Play pitch period contour
• Two-Band subband coder
• Phase Vocoder
• Isolated, speaker-trained, digit recognizer

51
 Speech Processing Exercises
• Search for MATLAB Central using local browser
• Click on file exchange; search for speech processing exercises
• Create a directory in which the various speech processing apps and data folders will
be placed (e.g., speech_apps); use the full path in the pathnew routine (see
below); e.g., C:\data\speech_apps
• If loading one or more speech processing exercises, download the following:
• Read_Me which explains the downloading process
• pathnew_matlab_central which links a set of functions and several data sets
to the current path including:
• runGUI – the set of files for creation of GUIs for the various speech apps
• speech_files – set of files used to demonstrate matlab apps capabilities
• functions_lrr – set of matlab functions used by the various matlab apps
• high pass filter signal – filtering function for speech signals
• isolated digit files – training and testing files for isolated digit recognition
• cepstral coefficients – used for vector quantization matlab app
• Be sure to put each of the downloaded folders in the chosen directory so that the
path links will be consistent
52
 Summary
• Set of about 60 MATLAB speech processing
exercises
• Exercises aligned with distinct sections in the
textbook TADSP by Rabiner/Schafer
• Each exercise has an associated Graphical User
Interface created using a GUI LITE program and
created expressly for these speech processing
exercises
• GUI LITE design and implementation Callbacks
are in totally separate code packages
53