ECE 531: Speech Signal Processing

Home : Courses :
ECE 531 - Speech Signal Processing

WWW Resources

Timetable Information

Engineering Courses on the Web (eCOW)

eCOW2 Courses

Edit This Page

Catalog Description

531 Speech Signal Processing. II, Irr.; 3 cr. Aerodynamic and acoustic mechanisms of sound production in speech. Multi-tube acoustic models of the vocal tract. Pitch detection, spectrographic analysis by Fourier and LPC methods. Speech synthesis, low bit rate speech coding, feature extraction for speech recognition. P: ECE 431 & Comp Sci 302.

Course Prerequisite(s)

ECE 431
Comp Sci 302

Prerequisite knowledge and/or skills

Write computer programs in an object-oriented language such as Java or C++,
Understand and apply OO class derivation, virtual object methods,
Able to apply linear-system concepts -- linear time-shift invariance, system functions, Laplace and z-transforms
Able to represent digital filters as difference equations, signal flow graphs, and as z-transform system functions,
Understand Fourier transform properties in continuous and discrete domains.

Textbook(s) and/or other required material

Recommended readings

L.R. Rabiner, R.W. Shafer, Digital Processing of Speech Signals

J.R Deller, Jr., J. G. Proakis, J. H. Hansen, Discrete-time Processing of Speech Signals

J.D. Markel, A.H. Gray, Jr., Linear Prediction of Speech

D. O'Shaughnessy, Speech Communication.

Course objectives

Develop an understanding of the relationship of vocal tract shapes and physical acoustics to the acoustic speech signal,
Use a spectrum analyser to relate the acoustic speech signal to acoustical processes,
Design and implement digital filters to synthesize speech and code speech at a low bit rate,
Implement speech analysis and speech synthesis modules using object-oriented software programs, using techniques such as class derivation, the use of software objects as components in a larger software system.

Topics covered

Object-oriented architecture for a software system for the acoustic analysis of speech,
Extending an acoustic analysis software system using OO techniques -- class derivation, Observer and Strategy design patterns,
The FFT algorithm, the real FFT and its implementation in software,
The application of spectrum analysis to the acoustic analysis of speech,
The source-filter model of speech production,
Implementation of the source-filter model with parallel 2nd-order filters,
Vocal tract shapes for vowels, acoustic equivalent circuits to vocal tract shapes, computing the vocal tract transfer function from an acoustic transmission line description,
Linear predictive coding (LPC) analysis: least-squares design of a serial-pole vocal-tract filter, the Levinson-Durbin algorithm for determining LPC coefficients, the relationship between the LPC vocal tract filter and an acoustic tube,
Pitch analysis based on waveform correlations, vocoders and waveform coders.

Class/laboratory schedule

Weeks 1 and 2 -- FFT, real FFT, implementation of a spectrum analyser in C++ using a GUI framework, spectrum analysis of speech,
Weeks 3 and 4 -- Source-filter model, implementation of a parallel-formant speech synthesizer, filters as software plug-in modules to speech analysis software, vowel categories,
Weeks 5 and 6 -- Geometric vocal tract model for vowels, acoustic transmission-line theory, computing the vocal tract acoustic transfer function using a transmission line model, persistent objects and visual application builders, Fant's nomogram and inflections in formant trajectories with articulatory movement,
Weeks 7 and 8 -- LPC analysis, LPC forward and inverse filters, Levinson-Durbin algorithm, relationship to acoustic tubes,
Weeks 9 and 10 -- Vocoders, pitch determination of speech using correlation methods.
Weeks 11-15 -- Student projects to be selected from alternate filter structures for parallel formant synthesis, vocoder synthesis with formant or pitch modifications, use of a parallel formant synthesizer to investigate perceptual confusion in vowel category, 1-D finite-element vocal tract acoustic model, predictive and subband low-bit rate speech coders.

Contribution of course to meeting the professional component

This course contributes primarily to the students' knowledge of engineering topics, but does not provide design experience.

The following statement indicates which of the following considerations are included in this course: economic, environmental, ethical, political, societal, health and safety, manufacturability, sustainability.

manufacturability

Relationship of course to undergraduate degree program objectives and outcomes

This course primarily serves students in the department. The information below describes how the course contributes to the undergraduate program objectives.

knowledge in the basic techniques of mathematics and the physical sciences strong skills in problem solving, leadership, and communication a capacity to use their engineering education to contribute to their communities the ability to make thoughtful, well-informed career choices

Assessment of student progress toward course objectives

Course grade based on 1) assigned projects involving both DSP design and computer program implementation in C++, 2) in-class mid-term examination, and 3) final course project regarding the implementation of a vocoder, waveform coder, or related application.

Person(s) who prepared this description

Paul H. Milenkovic

Copyright 2007 The Board of Regents of the University of Wisconsin System
Date last modified: 18-Jul-2007
Content by: ece@engr.wisc.edu
Accessibility

Web services
UW-Madison : COE : ECE : ECE Site Map