Com Sci 295 Digital Sound Modelling: course information

Goals of the Course

In this course we learn how the basic structure of sound perception affects the useful ways of processing sound through digital computations. The focus is on basic synthesis techniques, rather than on signal analysis, or on special applications of synthesis such as music or speech. The text title is misleading---it is not as closely oriented to music as it seems.

Prerequisites

Introductory computer programming (ComSci 105/106 or 110/111 or 115/116), basic knowledge of trigonometry, calculus, and complex numbers.

Electronic Communications

We make crucial use of three forms of electronic communication in CS221: shared files, the World Wide Web (WWW), and electronic mail.

• You must provide a publicly readable directory named CS295 in your home directory on the CS computer file system. I will collect links to these in my own public class directory: /stage/current/CS295/Class_members. Store all online materials that you develop for this class in your public directory, so that we can share work. I also encourage you to develop your own WWWeb materials.
• I will also provide course materials in a form accessible through WWW, at the URL http://www.classes.cs.uchicago.edu/classes/current/CS295/. The recommended tools for viewing this material are Mosaic and NetScape from graphics terminals and Lynx from character terminals.
• There is a public class discussion, consisting of comments attached by students and the instructor to various online course materials. All questions and comments that are not personal or confidential should be added to this discussion, so that everyone can benefit from them.
• Personal discussions, particularly about grading, and requests for appointments, may be addressed to me by electronic mail at odonnell@cs. These will normally be the only types of correspondence that should be made by electronic mail.

Topics

The sequence of topics in this course is organized around different techniques for synthesizing and/or manipulating sound, roughly increasing in sophistication and power.

• Continuous sound, sampling, and quantization. Sound is naturally represented as a continuous function from a real number representing time, to a real number representing the pressure of the air at a certain point, or perhaps the displacement. For some purposes, it seems better to regard sound as a continous function from time to a complex number, whose real part is some physical value such as pressure and whose imaginary part is a related value such as displacement. In either case, computer models of sound digitize the continuous function in two crucial ways: (1) they sample the function at a discrete set of input time values, and (2) they quantize the output pressure/displacement values into fixed-point or floating-point approximations. Each of these forms of digitization limits the fidelity with which we can represent a given sound.
• Naive sound processing An obvious approach to sound synthesis is to record real sounds, and manipulate them computationally. We study some simple computational ways of manipulating the speed, duration, and pitch of recorded sounds, and discover their limitations.
• Additive synthesis with enveloping Steady sounds may be constructed by adding simple periodic components, usually sine waves. Then, these steady sounds may be shaped by amplitude modulation with envelope functions that control the attack and decay. We discover what sorts of sounds can and cannot be constructed conveniently in this way.
• Filtering A filter is a sound transducer. We explore the control of sound by applying different sorts of excitation to a variety of filters.
• Special topics. If time permits, which it probably won't, we will explore some more specialized topics, such as frequency modulation, reverberation, or sonic imaging.

• [A.]Sound perception. We develop an abstract and approximate understanding of the way the ear analyzes vibrations of the air into component frequencies. We discover the shortcomings of overly simplistic descriptions of sound perception.
• [B.]Transform theory. We review the Fourier Transform as a mathematical tool for analyzing sound into component frequencies. We see how some of the shortcomings of the Fourier Transform are corrected in the Z Transform and the Wavelet Transform.
• [C.]Experiment. We make a lot of noises and listen to them. This is the fun part.

Class Work and Grading

I last taught this course in Spring 1999, but due to the unforseen, there may be some adjustment to the plans. I anticipate a modest amount of pencil-and-paper homework to exercise mathematical tools. There will probably be a midterm and final exam, but these will test only basic knowledge and skill, not creative problem solving. The main work for the course will be a project, involving the creation of interesting sounds such as a violin tone, a piano tone, or a sung vowel (one of the above, not all three). All project work will be shared around cooperatively among all students. I will grade your project, not on the grounds of the program(s) that you write, but on a 1-hour private interview in which you demonstrate and explain the results of the project. You may use anyone's results in your interview (with attribution, of course), but I will grade you on how well you explain the results in terms of the ideas discussed in class. The course grade will be determined mainly by the project interview and participation in the project. Homeworks and exams may count up to 40%, depending on how extensive and significant they turn out to be. I strongly recommend that you do a practice interview well before the end of the quarter. I will be available for practice interviews every week, but you must take the initiative to schedule one.

• Draft of lecture notes

odonnell@cs.uchicago.edu