ECE 537 Speech Processing

ECE 537 Schedule (Fall 2009)

L/W D Date Lecture and Assignment
1/1 M 8/24 Self-introductions from each student.
Lecture: Overview: Speech production, Acoustics, Psychophysics, neural and information processing.
Entropy: Its meaning, definition and the intuition behind it;
Read: Flanagan, Chapt. 1 djvu
2 W 8/26 Lecture: Mechanisms of Speech production: Sounds of speech: Vowels and consonants; Chapter 2 djvu;
HW1: IPA, Information Theory, and Basic Acoustics (djvu, due Wed 9/2/07, Solutions)
3 F 8/28 Lecture: Basic acoustics and the ABCD-Transmission (Chain) matrix; Acoustic Transmission lines;
Text: Flanagan, Ch. 3.1-3.2 djvu;
Read: Flanagan, Ch. 2.2 djvu
4/2 M 8/31 Lecture: Solution of 1-D transmission line equation:
Text:
5 W 9/2 Lecture: Intensity, speech power; sound level; dB, dB-SPL, Pressure, volume velocity, impedance. Vowels, Formants;
HW2: TL and reflectance (due Wed 9/16; (PDF)) (solutions)) (tubemodel.m))
Reading: Kelly and Lochbaum, 1962
6 F 9/4 Lecture:d'Alembert solutions for 1-D and 3-D transmission lines; Sound propagation in tubes in speech production; Introduction to reflectance; Conversion tables for 2-ports (djvu);
Read:: Model of Ear drum, Parent and Allen 2007 (pp 918-920) (djvu, pdf);
Review Matlab filter design, bilinear Z, FIR, IIR
-/3 M 9/7 Labor Day Holiday -- No class
7 W 9/9 Guest Lecture, Jont Allen: Transmission Lines with complex loads and the Propagated Reflectance;
Text: Flanagan, Ch. VI (Sect. 6.262, pp 272-276);
Readings: Bilbao PhD Thesis p. 1-15 djvu
8 F 9/11 Guest Lecture, Jont Allen:Impedance and reflectance at a tube junction (Karal correction djvu), and half-sphere (the mouth)
Text:: Ch. 3, Sec. 3.3, pages 136-152
9/4 M 9/14 Lecture: Spherically symmetric waves, radiation impedance, and radiation characteristic
10 W 9/16 Lecture: Radiation impedance
HW3 (pdf), due Wed 9/23/07, (solutions here).
11 F 9/18 Lecture: Glottal impedance; Thevenin and Norton equivalents; Bernoulli's equation Wiki
Text: Flanagan pp 41-53; Read: vanDen Berg (1957) djvu
12/5 M 9/21 Lecture: STFT window methods; Inverse STFT; STFT for speech processing with analysis/synthesis; filtering; Reading spectrograms
13 W 9/23 Lecture: STFT, inverse STFT.
HW4: Due October 14. PDF for HW4 and its solutions is here.. You will also want a copy of constants.m, code that defines the acoustic constants. Solutions: twoport.m, voicesource.m.
14 F 9/25 Lecture: Reflection coefficients and LPC
Reading: section 5.2 of the text, get the whole text here,, also see the LPC table.
15/6 M 9/28 Lecture: LPC normal equations
16 W 9/30 Lecture: formant frequencies, Levinson-Durbin recursion, line spectral frequencies.
HW5: Due October 7. PDF here. Because there is an exam coming up, I am already making the solutions available here; I recommend that you do the problems, then check them against the solutions before you hand in the assignment.
17 F 10/2 Lecture: Cepstral analysis;
Read:: Flanagan 5.3.
18/7 M 10/5 Lecture: Magnitude cepstrum and complex cepstrum; discrete-time cosine transform, discrete cosine transform; LPC cepstrum
19 W 10/7 Lecture: Review for Exam I
- - 10/8 Exam I: Modeling the VT, STFT, Signal processing of speech (Solutions Here)
Time and Place: Everitt 168, 7-9pm
Rules: 1 8.5x11 crib sheet, two sides, hand written. No computer print sheets!!; but note, you will be given a copy of the lpc table. Leave cell phones home. Calculators on the floor and off when you are not using them.
- F 10/9 no class due to Exam I
20/8 M 10/12 Lecture: Psychoacoustics I: Intensity JND and the near-miss;
Internal noise model of the JND Riesz pure-tone intensity JND (1928) djvu Masking;
Weber's and Fechner's Law;
Introduction to loudness, Steven's Law;
Loudness Lecture notes (Allen): pdf
Read:: Flanagan Chapter 4 (pdf, original pdf, djvu)
21 W 10/14 Lecture: Psychoacoustics II: Frequency JND, semitone, Internal noise and Masking; relation between the intensity and frequency JND (Cochlear frequency response and the slope of the tuning curve);
Read:: Fletcher and Munson (1933) (pp 82-94) djvu
HW6: Due October 21, PDF here, Solutions here.
22 F 10/16 Lecture: Guest lecture, James J. Johnston, Spatial Hearing
23/9 M 10/19 Lecture: Cochlear Physiology I:Middle ear and inner ear (Cochlear) anatomy, basilar membrane, 1D Models, Hair cells, Nonlinear basilar membrane;
Read:: Allen Review djvu
24 W 10/21 Lecture: autocorrelation and power spectrum of a sinusoid, and of broadband noise, in both discrete time and continuous time. Tones masking tones.
Read:: Review of Cochlear Modeling (pp 1-19) (djvu);
HW7: Due October 30, PDF here, Solutions here.
25 F 10/23 Lecture: Noise masking tones: narrowband noises, wideband noises. Equivalent rectangular bandwidth of the cochlear filter is about 1mm on the basilar membrane. Perceptual frequency scales based on ERB.
IHC Lecture Notes (djvu): Modeling the Cochlea and Organ of Corti, and
Read:: Wegel and Lane (1924), Part II (djvu)
26/10 M 10/26 No class
27 W 10/28 Lecture: Prof. Wickesberg, Auditory Pathway (AN+CN) ppt (Fall 2006 Part I: djvu, Part II: djvu)
28 F 10/30 Lecture: Pitch
HW8: (not due) PDF here, Solutions here
29/11 M 11/2 Lecture: Spatial hearing: ITD, IID, and HRTFs
30 W 11/4 Lecture: Review for Exam II;
- - 11/5 Exam II: Psychoacoustics, Physiology;
Rules: one 8.5x11 facts-sheet two sides (hand written), plus your Exam I facts-sheet
Time and Place: Everitt 168, 7-9pm
Solutions Here
- F 11/6 No Class because of Exam II.
31/12 M 11/9 Lecture: Room acoustics: Image source model, reverberation time.
Read:: Allen and Berkley, 1979, JASA 65(4):912-915, Image method for efficiently simulating small-room acoustics
32 W 11/11 Lecture: Psycholinguistics and semantics. Levelt's model. Spatial, propositional, and procedural knowledge. Entities, truth, and functions.
Read:: Language chapter. Altmann, Pierrehumbert, and Motter, PLoS 1, 2009.
HW9: Despite what it says on the assignment, HW9 is due Wednesday, December 2. (pdf here)
33 F 11/13 Lecture: Finite state transducer. Composition. Finite state models of Formulator, Encoder, Articulator, Channel, Perception, and Understanding. Maximum likelihood parameter estimation. Hidden Markov model.
Read:: Language chapter.
DUE: Preliminary version of your Final Exam to me for my review
34/13 M 11/16 Lecture: Information Theory I: Information. Entropy of a Markov source. Relative entropy. Conditional entropy of a Markov channel.
Read:: Shannon (1948) (djvu)
35 W 11/18 Lecture: Information Theory II: Differential entropy and conditional entropy of a Gaussian. The Gaussian channel. Mutual information. Channel capacity.
Read:: Shannon (1948) (djvu)
36 F 11/20 Lecture: Human speech recognition (HSR), Articulation Index (AI), average score: Pc(AI)=1-echanceeminAI, The confusion matrix (CM), maximum entropy syllable error models, etc.; Allen notes (djvu)
Read:: Continue with French and Steinberg (1947)djvu; Miller Nicely (1955) djvu; Miller Nicely confusions as a function of the articulation index; entropy, grouping and chance djvu
DUE: HW9
Assignment: Work on Final exam
- - - Thanksgiving Holiday (11/22-12/1)
37/14 M 11/30 Lecture: Effects of language and semantic context Miller (1962), Boothroyd (1988),
Allen Notes: Events and the AI(djvu)
Read: Miller, Heisen and Lichten (1951) (djvu)
38 W 12/2 Lecture: Language context models, Bronkhorst93.djvu
- F 12/4 No Lecture
39/15 M 12/7 Lecture: Semantics, Psycholinguistics, Historical Linguistics, Language in Practice, and Black Swans
Read: W. Li Random texts exhibit Zipf's-Law djvu
40 W 12/9 Last class: Final overview of the course material, re your final exam.
Free Pizza.
- R 12/10 Final Due on reading day, Dec 10.

Final

The final is a 20-25 page paper, written in the style and format (but single column) of a journal paper, that discusses everything that you have learned in this course. Writing style, spelling, figures, labels of figures, are all part of the grade.

Here is a final exam template file. You may use this file if you wish.

The final is graded based on a list of all the topics that are covered. The list of topics is provided in the final exam template file, above. If there is a paragraph that discusses each topic on my list, then you get at least 1 point, and if the discussion covers the topic effectively, you can get up to 5 points. There are at least 20 topics on the list. When you get to 100 points, you get an A+ on the exam. I expect that you draw on the homework as a starting point. Don't just dump the homework into the exam without modification, that won't get you points. Don't just dump a large number of unexplained figures (that you got from someone else for example) and expect to get points. I need words around each figure. I am looking for insightful comments that link the material together.

Your comments on the relevance of each of the topics I covered in this course, homework problems, exams, etc., are welcome. No points will be taken off, nor given, for strong opinions on my teaching style, or lack thereof, organization, or lack thereof, etc. Please put all such comments in a discussion section at the end of the paper, isolated from the rest of the material.