UCLA Dept. of Electrical Engineering EE214A: Digital Speech Processing Problem Set 7 (LAST ONE!) Due: 3/2/2011 Reading Assignment: Chapter 8.
1. Suppose that using LPC you estimated the transfer function to be:
H (z ) = 1=1 ; 2z ;1 ; 6z ;2 + z ;3 ; 2z ;4
UCLA Dept. of Electrical Engineering EE M214A, Winter 2011 Problem Set 4 Solution
February 7, 2011
1. 7.13 Answer:
1
2
2. 7.14 Answer:
3
4
5
3. 7.18 Answer:
6
4. Answer: Omitted. 5. (a) To avoid aliasing in the time domain, the number lters must be at lea
Midterm Solution
EE214A Winter 2016
Question 1: (15 points )
Pitch: A < B < C < D
Number of periods in the time domain:
Waveform 2 < Waveform 3 < Waveform 4 < Waveform 1
The space between adjacent harmonic frequencies:
Spectrum 3 < Spectrum 4 < Spectrum 2
UCLA
Dept. of Electrical Engineering
EE214A
Problem Set 1
Due: 01/13/2016
Reading Assignment: Chapters 1 and 2.
1. A signal xa (t) is band limited to 10 kHz and was sampled at 20 kHz. A DFT
of size N=1000 was then computed.
(a) What is the spacing between
Lec 10B: Sample Midterm Questions
Lec10B:SampleMidtermQuestions
Sample Problem 1
Consider a neutral vowel with formant frequencies at 500, 1500, 2500, 3500
Hz etc.,
Hz,
etc and a fundamental frequency (F0) of 200 Hz
Hz. The time waveform of
the vowel is s
2015/2/24
KernelFunctionsforMachineLearningApplications|~/cesarsouza/blog
KernelFunctionsforMachineLearningApplications
Inrecentyears,Kernelmethodshavereceivedmajorattention,particularlydueto
the increased popularity of the Support Vector Machines. Kernel
SPEAKER IDENTIFICATION SYSTEM
Yunxuan Yu, Zhuangtian Zhao
University of California, Los Angeles
Electronic Engineering Department, Computer Science Department
[email protected], [email protected]
ABSTRACT
In this report we introduce the background, app
ENHANCED SPEECH CLASSIFICATION AND
PITCH DETECTION
Peter Vepfekr and Michael S. Scordilisit
TElectrical and Electronic Engineering Department
The University of Melbourne, Australia
*Wire Communication Laboratory
The University of Patras, Greece
Abstract S
UCLA
Dept. of Electrical Engineering
EEEldA
Online IKIIlclteii'll Exaim 2012
This exam consists of five questions. Please SllOW all steps leading to your
answers. An accurate answer with no justification will not receive full credit.
The exam is 2 hours
Lecture 10
EE214A
Abeer Alwan
Midterm Review
Material covered: PS1 thru PS4, assigned book readings, lectures
1-10 (excluding 9), and other reading material posted. One cheat
sheet (8.5 x 11, double sided) is allowed. No calculators, phones,
or other elec
1/23/2014
Lecture 7
EE214A
Abeer Alwan
Mystery Spectrogram
Ul
Choose between: Enjoy, was, that, and, by, people,
little, simple, between, very,
those
1
1/23/2014
The Short Time Fourier Transform (STFT)
X n (e j ) x(m) w(n m)e jm
m
an ( ) jbn ( )
| X n (e
Lecture 5
Text to Speech Synthesis (TTS)
and Time Domain Processing
(some slides are courtesy of Prof.
Larry Rabiner)
1
Wavesurfer Demo
Download software from:
www.speech.kth.se/wavesurfer/download.html
Time and Frequency Domain Analysis: pitch
estimation
Lecture 3
EE 214A
Abeer Alwan
Last Time
Transforms
Properties of the DFT
LTI systems
Relationship between CT and DT
signals/transforms
Filtering
Up/Down Sampling
1
Today
Sounds of American English
Terminology used in speech processing
LTI model of speec
A spectraltemporal method for robust fundamental frequency
tracking
Stephen A. Zahoriana and Hongbing Hu
Department of Electrical and Computer Engineering, State University of New York at Binghamton,
Binghamton, New York 13902, USA
Received 14 December 20
UCLA
Dept. of Electrical Engineering
EE214A
Problem Set 5
Due: 2/22/2016
Reading Assignment: Rabiner and Schafer: Chapter 4 except for Section
4.5. Chapter 9 till Section 9.4.
1. Problem 4.3.
2. Problem 4.4.
3. Problem 9.1
4. Four consecutive samples of a
UCLA
Dept. of Electrical Engineering
EE214A (online)
Problem Set 2
Due: 01/25/2016
Reading Assignment: Chapter 3 and Chapter 5 (except for Sections 5.1.3,
5.1.4, 5.2.3, and 5.2.4).
1. Transcribe your rst and last name phonetically. Then, analyze the sound
UCLA
Dept. of Electrical Engineering
EE214A
Problem Set 3
Due: 2/01/2016
Reading Assignment: Rabiner and Schafer: Ch. 6 except Section 6.3.1.
1. Problem 6.7.
2. Problem 6.13.
3. Problem 6.16.
1
4. The conguration shown in the gure below is an idealized vo
Lecture 3
EE 214A
Abeer Alwan
Last Time
Transforms
Properties of the DFT
LTI systems
Relationship between CT and DT
signals/transforms
Filtering
Up/Down Sampling
1
Today
Sounds of American English
Terminology used in speech processing
LTI model of speec
Lecture 4
EE214A
Abeer Alwan
(some slides are from Prof. Rabiner)
Last Time
Classification of sounds, and mathematical
models of the source/excitation: function
with is voiced (with and without noise), or
voiceless. Voicing could be modeled as an
impulse
EE214A Lecture 2
Abeer Alwan
Last Time
Frequency Transforms for CT and DT signals:
X ( s), X ( j), ck ; X ( z ), X (e j ), X (k ), ck
Periodic signals have a line spectrum while
aperiodic signals have a continuous spectrum.
Also recall that:
/ T where T
EE214A Digital Speech Processing
Lecture 1
Abeer Alwan
all rights reserved
Denes and Pinson (1963)
1
Disciplines involved in Speech
Research
Engineering and Computer Science
Linguistics (phonetics, phonology, semantics,
syntax, etc.)
Psychology, Biolog
Lecture 5
Text to Speech Synthesis (TTS)
and Time Domain Processing
(some slides are courtesy of Prof.
Larry Rabiner)
1
Wavesurfer Demo
Download software from:
www.speech.kth.se/wavesurfer/download.html
Time and Frequency Domain Analysis: pitch
estimation
Lecture 6: The STFT
EE214A
Abeer Alwan
The Short Time Fourier Transform (STFT)
X n (e j ) x(m) w(n m)e jm
m
an ( ) jbn ( )
| X n (e j ) | e j n ( )
w(n-m) is a real window that determines the portion of the signal
to be analyzed. The STFT is sometimes wr
Introduction to Automatic
Speech
Recognition Systems (ASR)
Abeer Alwan
Speech Processing and Auditory Perception Laboratory
Speech Recognition System Overview
Model
Training
Speech
Signal
Acoustic
Model
Feature
Extraction
Text
Output
Decoding
Network
Dict
1/23/2014
Lecture 7
EE214A
Abeer Alwan
Mystery Spectrogram
Ul
Choose between: Enjoy, was, that, and, by, people,
little, simple, between, very,
those
1
1/23/2014
The Short Time Fourier Transform (STFT)
X n (e j ) x(m) w(n m)e jm
m
an ( ) jbn ( )
| X n (e
Lecture 14
EE214A
Abeer Alwan
Recall the LTI Model of Speech Production
1
Analysis Techniques Studies Thus
Far
We have studied how to analyze the speech
signal using time-domain features and by
using the STFT and LPC
The STFT provides temporal-spectral
Lecture 8
EE214A
Abeer Alwan
Signal Reconstruction
Filterbank Summation (FBS)
Overlap and Add (OLA)
1
Recall that the Short Time Fourier Transform
(STFT) is defined as:
X n (e j ) x(m) w(n m)e jm
m
an ( ) jbn ( )
| X n (e j ) | e j n ( )
Second Interpr
Lecture 13
EE214A
Abeer Alwan
Speech Application
p
In speech, we assume
s(n) ak s(n k ) Gu (n)
Transfer function is
S ( z)
U ( z)
k 1
G
p
1 ak z k
G
A( z )
k 1
Goal: Estimate aks
p
Assume prediction signal is
~
s ( n ) ak s ( n k )
k 1
Prediction err
Lecture 12
EE214A
Abeer Alwan
Some Applications
LPC10 speech coder
The SIFT algorithm
1
Application Example:
Pitch-Excited LPC Transmitter
Block Diagram of a
Pitch-Excited LPC Receiver
2
Direct Form Implementation of the
Vocal Tract Filter
Speech Coding
EE 214A
Acoustic Wave Equations and Speech Production Models
For a lossless tube of uniform area (A), the one-dimensional wave equations are:
@ 2 p(xt)
@x2
=
2
1 @ p(xt)
c2 @t2
@ 2 u(xt)
@x2
=
@ 2 u(xt)
c @t2
1
2
where p(x t) u(x t) are the pressure and v
train the system in quiet, then test it in quiet, then train it in quiet, then test it in noise
use 6 sample to train, 2 to test.
statistical model is fixed, you need to play with
the input parameter
UCLA
Dept. of Electrical Engineering
EE214A
Project Des
f0 is from the vocal transform, has nothing to do with the format
Lecture 6: The STFT
EE214A
Abeer Alwan
Because the signal is time variant, it varies with time significantly, so we should only take part of it to do the analysis.
The Short Time Fourier Tr
Lecture 4
EE214A
Abeer Alwan
(some slides are from Prof. Rabiner)
Last Time
Classification of sounds, and mathematical
models of the source/excitation: function
with is voiced (with and without noise), or
voiceless. Voicing could be modeled as an
impulse
Department of Electrical Engineering, UCLA
EE214A - Winter Quarter 2015
Homework 2 Solutions
1.
2.
3
3.4 The sounds in the word /and/ are /AE/ /N/ /D/ and the (extremely) approximate locations of the sounds
are:
/AE/ samples 400-3800
/N/ samples 3800-54