RAJALAKSHMI INSTITUTE OF TECHNOLOGY

DEPARTMENT OF ELECTRONICS AND COMMUNICATION

ENGINEERING

CEC356-SPEECH PROCESSING LAB MANUAL

(2024-25-ODD)

FACULTY INCHARGE: DR.G. NIRMALA PRIYA, MS. ANCY SOPHIA

Program 1: Write a MATLAB program that reads a speech file and plots the waveform,
spectrum and autocorrelation sequence of any three voiced segments present in the given
speech signal.
Aim:
To Write a MATLAB program that reads a speech file and plots the waveform, spectrum
and autocorrelation sequence of any three voiced segments present in the given speech
signal.
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
PROGRAM:
clear all;
fp=fopen ('c:\a.wav', 'r');
fseek (fp, 21044,-1);
a=fread (fp, 8000);
a-a-128;
plot (a);
xlabel('sample number');
ylabel('amplitude');
title('plot of the utterance i');
Fs = 22100;
t = 0:1/Fs:.296;
h=spectrum.welch;
d = psd (h, a, 'Fs', Fs);
figure; % Calculate the PSD
plot (d);
to find pitch period of the signal using spectral autocorrelation of voiced speech
clear all;
fp=fopen('watermark.wav');
fseek (fp, 224000,-1);
a=fread (fp, 2048);
subplot (2,1,1);
plot (a);
title('plot of voiced part of a signal');
xlabel('sample number');
ylabel('amplitude');
a=a-128;
b=abs (fft (a));
f=22100/2048:22100/2048:22100;
subplot (2,1,2);
plot (f,b);
title('plot of FFT of a signal');
xlabel('frequency');
ylabel('amplitude');
for k=1:400,
sum (k)=0;
end
for k=1:400,
for i=1:45,
sum (k) =sum (k) + (b (i) *b (i+k));
sum (k) =sum (k)/52000;
end
end
figure;
plot (sum);
title('plot of spectral correlation of voiced signal');
xlabel('sample number');
ylabel('correlation');

Output
Program 2: Write a MATLAB program that reads a speech file and plots the waveform,
spectrum and autocorrelation sequence of any three unvoiced segments present in the
given speech signal.
Aim:
To Write a MATLAB program that reads a speech file and plots the waveform, spectrum
and autocorrelation sequence of any three unvoiced segments present in the given speech
signal.
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
Program:
clear all;
fp=fopen ('c:\a.wav', 'r');
fseek (fp, 21044,-1);
a=fread (fp, 8000);
a-a-128;
plot (a);
xlabel('sample number');
ylabel('amplitude');
title('plot of the utterance h');
Fs = 22100;
t = 0:1/Fs:.296;
h=spectrum.welch;
d = psd (h, a, 'Fs', Fs);
figure; % Calculate the PSD
plot (d);
to find pitch period of the signal using spectrum of unvoiced speech
clear all;
fp=fopen('c:/h.wav');
fseek (fp, 38044,-1);
a=fread (fp, 2048);
subplot (2,1,1); plot(a);
title('plot of unvoiced part of a signal');
xlabel('sample number');
ylabel('amplitude');
a-a-128;
b=abs (fft (a));
c=log (b^2);
f = 22100/2048:22100/2048:22100;
subplot (2,1,2);
plot (f, c);
title('plot of FFT of a signal');
xlabel('frequency');
ylabel('amplitude');
disp (b);
for k=1:400, sum (k) =0; end
for k=1:400,
for i=1:45,
sum (k) =sum (k) + (b (i) *b (i+k));
sum (k) =sum (k)/52000;
end
end
figure;
plot (sum);
title('plot of spectral correlation of unvoiced signal');
xlabel('sample number');
ylabel('correlation');
Program 3: Write a MATLAB program that reads a speech file and plots the waveform,
spectrum and autocorrelation sequence of any three silence regions present in the given
speech signal.

Program 4: Write a MATLAB program to classify voiced and unvoiced segment

using various time domain measures
Aim:
To write a MATLAB program to classify voiced and unvoiced segment using various time
domain measures.
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
PROGRAM:
windowLength = round(0.03*fs);
overlapLength = round(0.025*fs);

f0 =
pitch(audioIn,fs,WindowLength=windowLength,OverlapLength=overlapLength,Range=
[50,250]);

figure
tiledlayout(2,1)

nexttile()
plot(timeVector,audioIn)
axis([(110e3/fs) (135e3/fs) -1 1])
ylabel("Amplitude")
xlabel("Time (s)")
title("Utterance - Two")
nexttile()
timeVectorPitch = linspace(twoStart/fs,twoStop/fs,numel(f0));
plot(timeVectorPitch,f0,"*")
axis([(110e3/fs) (135e3/fs) min(f0) max(f0)])
ylabel("Pitch (Hz)")
xlabel("Time (s)")
title("Pitch Contour")

OUTPUT:

PROGRAM 5: Write a MATLAB program to calculate MFCC for a Speech Signal

Mel-Frequency Cepstrum Coefficients (MFCC)
MFCC are popular features extracted from speech signals for use in recognition tasks. In
the source-filter model of speech, MFCC are understood to represent the filter (vocal
tract). The frequency response of the vocal tract is relatively smooth, whereas the source
of voiced speech can be modeled as an impulse train. The result is that the vocal tract can
be estimated by the spectral envelope of a speech segment.
The motivating idea of MFCC is to compress information about the vocal tract (smoothed
spectrum) into a small number of coefficients based on an understanding of the cochlea.
Although there is no hard standard for calculating MFCC, the basic steps are outlined by
the diagram.
AIM:
To write a MATLAB program to calculate MFCC for a speech signal.
SOFTWARE REQUIRED:
MATLAB
HARDWARE REQUIRED:
Personal PC, Microphones and Speakers
PROGRAM:
% Ensure you have MATLAB's Audio Toolbox installed to run this script
% Load a speech signal
[audioIn, fs] = audioread('speech.wav'); % Replace 'speech.wav' with your audio file
% Parameters for MFCC
numCoeffs = 13; % Number of MFCC coefficients
winLength = 0.025; % Window length in seconds
overlapLength = 0.015; % Overlap length in seconds
numFilters = 26; % Number of Mel filters

% Convert lengths from seconds to samples

winLengthSamples = round(winLength * fs);
overlapLengthSamples = round(overlapLength * fs);

% Compute MFCC
coeffs = mfcc(audioIn, fs, 'NumCoeffs', numCoeffs, ...
'WindowLength', winLengthSamples, ...
'OverlapLength', overlapLengthSamples, ...
'NumFilters', numFilters);

% Plot the MFCC

figure;
imagesc(coeffs');
axis xy;
title('MFCC');
xlabel('Frame Index');
ylabel('Coefficient Index');
colorbar;

PROGRAM 6: Implement ITU-T G.722 Speech Encoder in MATLAB

AIM: To write a MATLAB program to implement ITU-T G.722 speech encoder.
SOFTWARE REQUIRED:
MATLAB
HARDWARE REQUIRED:
Personal computer, Microphones and Speakers.
PROCEDURE:
ITU-T G.722 is a standard for high-quality audio coding, specifically for speech. It includes
a 7 kHz audio codec operating at 48, 56, and 64 kbit/s. Implementing the full G.722
encoder is complex, involving sub-band adaptive differential pulse code modulation (SB-
ADPCM).
PROGRAM:
% Simplified ITU-T G.722 Encoder in MATLAB
% Note: This is a simplified version and does not cover all the complexities of G.722.

% Read an input audio file

[audioIn, fs] = audioread('speech.wav'); % Replace 'speech.wav' with your audio file

% Check if the input audio is mono, if stereo convert to mono

if size(audioIn, 2) == 2
audioIn = mean(audioIn, 2);
end

% Parameters
fs_new = 16000; % New sampling rate for G.722
N = 2; % Decimation factor

% Low-pass and high-pass filters (simple example filters)

[b_low, a_low] = butter(6, 1/N, 'low');
[b_high, a_high] = butter(6, 1/N, 'high');

% Split signal into two sub-bands

low_band = filter(b_low, a_low, audioIn);
high_band = filter(b_high, a_high, audioIn);
% Down-sample the sub-bands
low_band_ds = downsample(low_band, N);
high_band_ds = downsample(high_band, N);

% ADPCM encoder for the low band (simple example, not actual G.722 ADPCM)
encoded_low = adpcm_encode(low_band_ds);

% ADPCM encoder for the high band (simple example, not actual G.722 ADPCM)
encoded_high = adpcm_encode(high_band_ds);

% Combine encoded sub-bands (this is simplified and not the actual G.722 bitstream)
encoded_signal = [encoded_low; encoded_high];

% Save the encoded signal (for demonstration purposes)

save('encoded_signal.mat', 'encoded_signal');

% Function for ADPCM encoding (simple example, not actual G.722 ADPCM)
function encoded = adpcm_encode(x)
% Initialize variables
x = x(:);
n = length(x);
encoded = zeros(n, 1);
predsample = 0;
step_size = 1;

for i = 1:n
diff = x(i) - predsample;
if diff >= 0
encoded(i) = 1;
else
encoded(i) = 0;
end
predsample = predsample + step_size * (2 * encoded(i) - 1);
end
end

PROGRAM 7: Write a MATLAB program to implement Wiener Filters for Noise

Reduction
AIM: To write a MATLAB program to implement Wiener filters for noise reduction.
SOFTWARE REQUIRED:
MATLAB
HARDWARE REQUIRED:
Personal computer, Microphones and speakers.
PROCEDURE:
Implementing Wiener filtering for noise reduction in MATLAB involves creating a Wiener
filter, which is an optimal filter for reducing noise in a signal. Here is a step-by-step
MATLAB program to achieve this:
1. Read the Input Signal: Load or generate the noisy signal.
2. Wiener Filter Design: Design the Wiener filter based on the noisy signal.
3. Apply the Filter: Apply the Wiener filter to the noisy signal to obtain the denoised
signal.
4. Plot the Results: Compare the original, noisy, and filtered signals.

PROGRAM:

% Wiener Filter for Noise Reduction

% Step 1: Read or Generate the Noisy Signal

fs = 1000; % Sampling frequency
t = 0:1/fs:1-1/fs; % Time vector
originalSignal = sin(2*pi*50*t) + sin(2*pi*120*t); % Original signal (sum of two
sinusoids)
noise = 0.5 * randn(size(t)); % Gaussian noise
noisySignal = originalSignal + noise; % Noisy signal

% Step 2: Wiener Filter Design

% Wiener filter needs the power spectral density (PSD) of the noise and the signal
% For simplicity, let's assume we know the PSD of the noise and the signal
signalPower = var(originalSignal);
noisePower = var(noise);

% Wiener filter design

snr = signalPower / noisePower;
H_wiener = @(f) (snr / (snr + 1));

% Frequency response of the Wiener filter

nfft = 1024; % Number of FFT points
f = linspace(0, fs/2, nfft/2+1); % Frequency vector
H_w = arrayfun(H_wiener, f);

% Step 3: Apply the Filter

% Fourier Transform of the noisy signal
NoisySignalFFT = fft(noisySignal, nfft);
NoisySignalFFT = NoisySignalFFT(1:nfft/2+1);
% Apply Wiener filter in the frequency domain
FilteredSignalFFT = NoisySignalFFT .* H_w;

% Inverse Fourier Transform to get the filtered signal

FilteredSignal = ifft([FilteredSignalFFT, conj(FilteredSignalFFT(end-1:-1:2))], nfft);
FilteredSignal = real(FilteredSignal(1:length(t))); % Take the real part

% Step 4: Plot the Results

figure;
subplot(3, 1, 1);
plot(t, originalSignal);
title('Original Signal');
xlabel('Time (s)');
ylabel('Amplitude');

subplot(3, 1, 2);
plot(t, noisySignal);
title('Noisy Signal');
xlabel('Time (s)');
ylabel('Amplitude');

subplot(3, 1, 3);
plot(t, FilteredSignal);
title('Filtered Signal using Wiener Filter');
xlabel('Time (s)');
ylabel('Amplitude');

% Display the SNR improvement

snrBefore = 10 * log10(signalPower / noisePower);
snrAfter = 10 * log10(var(FilteredSignal) / var(FilteredSignal - originalSignal));
disp(['SNR Before Filtering: ', num2str(snrBefore), ' dB']);
disp(['SNR After Filtering: ', num2str(snrAfter), ' dB']);

PROGRAM 8: Write a MATLAB program to find PLP coefficients for 256 samples of
voiced speech
Aim: To write a MATLAB program to find PLP coefficients for 256 samples of voiced
speech
SOFTWARE REQUIRED:
MATLAB
HARDWARE REQUIRED:
Personal computer, Microphones and speakers.
Program:
program to find PLP coefficients for 256 samples of voiced speech
clear all;
fp=fopen('watermark.wav');
fseek (fp, 224000,-1);
a=fread (fp, 256);
plot 256 points of voiced speech
subplot (2,1,1);
plot (a);
title('plot of voiced part of a signal'); xlabel('sample number');
ylabel('amplitude'); %find 256 point FFT
b=fft (a);
bl= (abs (b));
for i=1:256,
bl (i) =b1 (i) *bl (i);
calculation of squared power
end
calculate frequency in Hz for every FFT bin
for i=1:128,
end
f(i)=22100/256*i;
for i=1:128,
c1 (i)=b1 (i);
end
calculate Bark scale frequency for each frequency in Hz corresponding to FFT bin
for i=1:128,
bark (i)=6*log (2*pi*f (i)/1200*pi+sqrt ( (2*pi*f (i)/1200*pi)^2+1));
end
plot spectrum in Bark scale for each FFT bin. And find cube root of power spectrum bin
and plot it on the Bark scale.
for i=1:128,
end
c11 (i) =nthroot (c1 (i),3);
subplot (2,1,2);
stem (bark, c11);
title('plot of cube root of power spectrum in Bark scale for voiced speech');
xlabel('Frequency in Bark scale'); ylabel('Amplitude in dB');
for j=1:28,
sum (j)=0;
for i=1:58,
if ((bark (i)>10+ (j-1)*2.5) && (bark (i) <15+ (j-1)*2.5))
if (bark (i) <12.5+ (j-1)*2.5)
g(i)=((bark (i)-(2.5+2.5* (j-1)))*1/5); else
g(i)=((15+2.5* (j-1) -bark (i))*1/5);
end
sum (j)=sum (j)+c11 (i) *g (i);
end
end
end
Find IFFT of the resulting integrated values considering it as a signal.
d=ifft (sum); dl-real (d);
for i=1:14,
end
x (i) =d1 (i);
plot the first 14 MFCC coefficients by cepstral truncation.
stem (x);
title('plot of PLP coefficients for voiced speech');
xlabel('Frequency in mel scale');
ylabel('Amplitude in dB');

Output:

PROGRAM 9: Write a MATLAB Program to implement Combo Filters for Noise

Reduction

PROGRAM 10: Design a Speech emotion recognition system using DCT in MATLAB
AIM:
To design a speech emotion recognition system using DCT in MATLAB.
SOFTWARE REQUIRED:
MATLAB
HARDWARE REQUIRED:
Personal computer, Microphones and speakers.
PROCEDURE:

Designing a Speech Emotion Recognition (SER) system using Discrete Cosine Transform
(DCT) in MATLAB involves several steps, including preprocessing, feature extraction,
and classification. Here’s a simplified outline of how this can be implemented:

1. Preprocessing: Load and preprocess the speech signals.

 2. Feature Extraction: Extract features using DCT.
3. Feature Selection: Combine the features from DCT.
4. Classification: Train a classifier and test the system.

PROGRAM:
% Load necessary libraries
clc; clear; close all;
% Parameters
fs = 16000; % Sampling frequency
frame_size = 0.025; % Frame size in seconds
frame_stride = 0.01; % Frame stride in seconds
% Load the speech signal
[audio, fs] = audioread('emotion_speech.wav');
% Normalize the signal
audio = audio / max(abs(audio));
% Frame the signal
frame_length = round(frame_size * fs);
frame_step = round(frame_stride * fs);
frames = buffer(audio, frame_length, frame_length - frame_step, 'nodelay');
% Apply Hamming window
hamming_window = hamming(frame_length);
frames = frames .* hamming_window;

% Step 2: Feature Extraction

% DCT Features
dct_features = dct(frames);

% Step 4: Classification
% Split data into training and testing sets (using dummy data for example)
labels = [ones(1, size(frames, 2)/2), zeros(1, size(frames, 2)/2)]; % Dummy labels
train_features = features(:, 1:end/2);
test_features = features(:, end/2+1:end);
train_labels = labels(1:end/2);
test_labels = labels(end/2+1:end);

% Train a classifier (using k-NN for simplicity)

mdl = fitcknn(train_features', train_labels);

% Test the classifier

predicted_labels = predict(mdl, test_features');

% Evaluate the classifier

accuracy = sum(predicted_labels' == test_labels) / length(test_labels);
disp(['Classification Accuracy: ', num2str(accuracy * 100), '%']);
PROGRAM 11: Design a Speech emotion recognition system using WPT in MATLAB
AIM:
To design a speech emotion recognition system using WPT in MATLAB.
SOFTWARE REQUIRED:
MATLAB
HARDWARE REQUIRED:
Personal computer, Microphones and speakers.
PROCEDURE:

Designing a Speech Emotion Recognition (SER) system using Wavelet Packet Transform
(WPT) in MATLAB involves several steps, including preprocessing, feature extraction,
and classification. Here’s a simplified outline of how this can be implemented:

1.Preprocessing: Load and preprocess the speech signals.

2.Feature Extraction: Extract features using WPT.
3.Feature Selection: Combine the features from WPT.
4.Classification: Train a classifier and test the system.

PROGRAM:
% Load necessary libraries
clc; clear; close all;
% Parameters
fs = 16000; % Sampling frequency
frame_size = 0.025; % Frame size in seconds
frame_stride = 0.01; % Frame stride in seconds
% Load the speech signal
[audio, fs] = audioread('emotion_speech.wav');
% Normalize the signal
audio = audio / max(abs(audio));
% Frame the signal
frame_length = round(frame_size * fs);
frame_step = round(frame_stride * fs);
frames = buffer(audio, frame_length, frame_length - frame_step, 'nodelay');
% Apply Hamming window
hamming_window = hamming(frame_length);
frames = frames .* hamming_window;

% WPT Features
wpt = wpdec(audio, 3, 'db1'); % 3-level wavelet packet decomposition with 'db1'
wavelet
wpt_features = wprcoef(wpt, [3 0]); % Extract coefficients at level 3

% Step 4: Classification
% Split data into training and testing sets (using dummy data for example)
labels = [ones(1, size(frames, 2)/2), zeros(1, size(frames, 2)/2)]; % Dummy labels
train_features = features(:, 1:end/2);
test_features = features(:, end/2+1:end);
train_labels = labels(1:end/2);
test_labels = labels(end/2+1:end);

% Train a classifier (using k-NN for simplicity)

mdl = fitcknn(train_features', train_labels);

% Test the classifier

predicted_labels = predict(mdl, test_features');

% Evaluate the classifier

accuracy = sum(predicted_labels' == test_labels) / length(test_labels);
disp(['Classification Accuracy: ', num2str(accuracy * 100), '%']);

PROGRAM 12: Write a MATLAB program to calculate positive and negative ZCR for
voiced and unvoiced segments
Aim:
To write a MATLAB program to calculate positive and negative ZCR for voiced and
unvoiced segments
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
PROGRAM:
clear all;
fp-fopen('watermark.wav'); fseek (fp, 60044,-1); a=fread (fp, 10000);
plot (a);
xlabel('sample number'); ylabel('amplitude');
title('plot of speech segment for 10000 samples');
file read and plotted. We are reading first 10000 samples after fseek by 60000 samples.
x=0; figure;
for i=5000:6999,
if (a(i)>128) && (a (i+1)<128)
x=x+1;
else x-x;
end
end
disp('number of zero crossings for voiced segment=');
disp (x);
samples between 5000 to 7000 are read and number of zero cross- ings are found.
fseek (fp, 65044,-1);
b=fread (fp, 2000);
plot (b);
xlabel('sample number');
ylabel('amplitude');
title('plot of voiced speech segment');
We are reading samples between 5000 to 7000 and are plotted. This is unvoiced
segment.
figure;
for i=7000:8999,
if (a(i)>128) && (a (i+1) <128)
x=x+1;
else x=x;
end
end
disp ('number of zero crossings for unvoiced segment='); disp (x);
samples between 7000 to 9000 are read and number of zero cross- ings are found.
fseek (fp, 67044,-1);
c-fread (fp, 2000);
xlabel('sample number');
ylabel('amplitude');
title('plot of unvoiced speech segment');
plot (c);
We are reading samples between 7000 to 9000 and are plotted. This is unvoiced
segment.
Output
number of negative zero crossings for voiced segment= 135
number of negative zero crossings for unvoiced segment=193

PROGRAM 13: Write a MATLAB Program to find Pitch Period of the Voiced Signal
Using Cepstrum Of Speech signal
Aim:
To write a MATLAB Program to find Pitch Period of the Voiced Signal Using Cepstrum of
Speech signal
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
PROGRAM:
to find pitch period of the voiced signal using cepstrum of speech signal
clear all;
fp=fopen('watermark.wav');
fseek (fp, 224000,-1);
a=fread (fp, 2048);
a=a-128;
subplot (2,1,1);
plot (a);
title('plot of voiced part of a signal');
xlabel('sample number');
ylabel('amplitude');
%2048 samples of the voiced signal are displayed
al=window (@Hamming, 2048);
subplot (2,1,2);
plot (al);
title('plot of Hamming window signal');
xlabel('sample number');
ylabel('amplitude');
for i=1:2048,
end
all (i) a (i) *al (i);
figure;
subplot (2,1,1);
plot (all);
title('plot of windowed voiced part of a signal');
xlabel('sample number');
ylabel ('amplitude');
the signal is passed via Hamming window and displayed
b=abs (fft (a11));
f
22100/2048:22100/2048:22100;
subplot (2,1,2);
plot (f,b);
title('plot of FFT of windowed signal');
xlabel('frequency');
ylabel ('amplitude');
c=log (b);
figure;
for i=1:1024,
d(i)=c(i);
end
f = 22100/2048:22100/2048: 11050;
subplot (2,1,1);
plot (f, d);
title('log spectrum of windowed voiced speech signal'); xlabel('Frequency in Hz');
ylabel ('amplitude in dB');
e-abs (ifft (c));
subplot (2,1,2);
plot (e);
title('cepstrum of voiced speech signal');
xlabel('Quefrency');
ylabel('amplitude in dB');
for i=1:80,
end
h(i)=0;
for i-81:1967,
end
h(i) =e(i);
for i 1968:2048,
h(i)=0;
end
figure;
subplot (2,1,1);
plot (e);
title('cepstrum of voiced speech signal after passing through
window' );
xlabel('Quefrency');
ylabel('amplitude in dB');
g=abs (fft (h));
for i=1:1024,
end
k(i)=g(i);
f = 22100/2048:22100/2048:11050;
subplot (2,1,2);
plot (f, k);
title('spectrum of voiced speech cepstral windowed');
xlabel('frequency');
ylabel('amplitude in dB');
disp (k);

PROGRAM:14: Write a MATLAB Code to find Formants from the Voiced Signal Using
Cepstrum of Speech signal
Aim:
To write a MATLAB Program to find Pitch Period of the Voiced Signal Using Cepstrum of
Speech signal
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
PROGRAM:
to find formants from the voiced signal using cepstrum of speech signal
clear all;
fp=fopen('watermark.wav');
fseek (fp, 224000,-1);
a=fread (fp, 2048);
a=a-128;
subplot (2,1,1);
plot (a);
title('plot of voiced part of a signal');
xlabel('sample number');
ylabel('amplitude');
$2048 samples of the voiced signal are displayed
al=window (@Hamming, 2048);
subplot (2,1,2);
plot (al);
title('plot of Hamming window signal'); xlabel('sample number');
ylabel('amplitude');
for i=1:2048,
end
all (i) a (i) *al (i);
figure;
subplot (2,1,1);
plot (all);
title('plot of windowed voiced part of a signal'); xlabel('sample number');
ylabel('amplitude');
the signal is passed via Hamming window and displayed b=abs (fft (all));
f = 22100/2048:22100/2048:22100;
subplot (2,1,2);
plot (f,b);
title('plot of FFT of windowed signal');
xlabel('frequency');
ylabel('amplitude');
c=log (b);
figure;
for i=1:1024,
d(i)=c(i);
end
f = 22100/2048:22100/2048:11050;
subplot (2,1,1);
plot (f, d);
title('log spectrum of windowed voiced speech signal'); xlabel('Frequency in Hz');
ylabel('amplitude in dB'); e-abs (ifft (c));
subplot (2,1,2); plot (e);
title('cepstrum of voiced speech signal');
xlabel('Quefrency'); ylabel('amplitude in dB');
for i=1:40,
h(i)=e(i);
end
for i=41:2007,
h(i)=0;
end
for i=2008:2048,
h(i)=e (i);
end
figure;
subplot (2,1,1);
plot (h);
title('cepstrum of voiced speech signal after passing through
window' );
xlabel('Quefrency');
ylabel('amplitude in dB');

PROGRAM:15: Write a MATLAB Code to find Formants from the Unvoiced Signal
Using Cepstrum of Speech signal
Aim:
To write a MATLAB Program to find Pitch Period of the Unvoiced Signal Using Cepstrum
of Speech signal
Software Required:
MATLAB
Hardware Required:
Personal PC, Microphones and Speakers
PROGRAM:
%to find formants from the unvoiced signal using cepstrum of a speech signal
clear all;
fp=fopen('watermark.wav');
fseek (fp, 23044,-1);
a=fread (fp, 2048); a=a-128;
subplot (2,1,1);
plot (a);
title('plot of unvoiced part of a signal');
xlabel('sample number');
ylabel('amplitude');
%2048 samples of the voiced signal are displayed
a1=window (@Hamming, 2048);
subplot (2,1,2);
plot (al);
title('plot of Hamming window signal');
xlabel('sample number');
ylabel('amplitude');
for i=1:2048,
a11 (i)= a (i) *al (i);
end
figure;
subplot (2,1,1);
plot (all);
title('plot of windowed unvoiced part of a signal'); xlabel('sample number');
ylabel('amplitude');
%the signal is passed via Hamming window and displayed b=abs (fft (all));
f = 22100/2048:22100/2048:22100;
subplot (2,1,2);
plot (f,b);
title('plot of FFT of windowed signal');
xlabel('frequency');
ylabel('amplitude');
c=log (b);
figure;
for i=1:1024,
end
d(i)=c(i);
f = 22100/2048:22100/2048:11050;
subplot (2,1,1);
plot (f, d);
title('log spectrum of windowed unvoiced speech signal');
xlabel('Frequency in Hz');
ylabel('amplitude in dB'); e=abs (ifft (c));
subplot (2,1,2);
plot (e);
title('cepstrum of unvoiced speech signal');
xlabel('Quefrency');
ylabel('amplitude in dB');
for i=1:40,
h(i) =e(i);
end
for i=41:2007,
h(i)=0;
end
for i=2008:2048,
h(i) =e (i);
end
figure;
subplot (2,1,1);
plot (h);
title('cepstrum of voiced speech signal after passing through window' );
xlabel('Quefrency');
ylabel('amplitude in dB');
g=abs (fft (h));
for i=1:1024,
end
k(i)=g(i);
f = 22100/2048:22100/2048:11050;
subplot (2,1,2);
plot (f, k);
title('spectrum of voiced speech cepstral windowed');
xlabel('frequency');
ylabel('amplitude in dB');
disp (k);