EEL6586: HW#4

Get PDF file

EEL 6586: HW#4

Assignment is due Friday, February 29, 2008 in class. Late homework loses $e^{\char93  of days late} -1$ percentage points. This assignment includes both matlab and textbook questions. This assignment must be completed in teams. Only one assignment should be turned in with all the names of the team members written on the front. The teams have been arbitrarily assigned as follows:

1. Arvind Iyer and Yan Yang
2. Kiran Kalidindi and Yu Fan
3. Salil Bibikar and Steven Spalding
4. Avantika Vardhan and Steven Wright
5. Karthik Talloju and Jiang Lu
6. Surbhi Singhal and Justin Zito
7. Padmanabhan Ramakrishnan and Yuelu Liu
8. Rahul Chaithanya and Oluwatosin Adeladan
9. Reno Varghese and Sarah Keen
10. Siddharth Gaddam and Ming Xue
11. Keshav Rajan, Duroseme Taylor and Eric Graves

Teammates should work together on each of the problems and agree on their solutions. As in real life, dysfunctional teams will be severely penalized. PART A: Thought Problems

A1

All other factors being equal, do you expect speaker-independent ASR systems to have a lower error rate than speaker-dependent ASR systems? Explain.

A2

Briefly explain how linguistic constraints can improve a speech recognition system.

A3

Class $\omega_1$ points are:

$$\left[ \begin{array}{c} -1 \\ -1 \\ +1 \end{array} ... ...[ \begin{array}{c} +1 \\ -1 \\ -1 \end{array} \right]$$

Class $\omega_2$ points are:

$$\left[ \begin{array}{c} +1 \\ +1 \\ -1 \end{array} ... ...[ \begin{array}{c} -1 \\ +1 \\ +1 \end{array} \right]$$

Find any weight vector $w$ such that $w^Tx>0$ for all class $\omega_1$ points and $w^Tx<0$ for all class $\omega_2$ points. Justify your answer.

A4

Using the points in [A3] as the training set, classify $[1,1,1]^T$ using 3-Nearest Neighbor voting. Remember that in k-NN classification, the k nearest neighbors of the point are found and the most common class label is used for classification.

A5

Consider the following sample points: The samples from class 1 are: $\left[\begin{array}{c} 0 0 \end{array} \right] \left[\begin{array}{c} 1 1 \end{array} \right] \left[\begin{array}{c} 2 1 \end{array} \right]$ The samples from class 2 are: $\left[ \begin{array}{c} 1 0 \end{array} \right] \left[ \begin{array}{c} 1 ... ... -1 \end{array} \right] \left[ \begin{array}{c} 1 -2 \end{array} \right]$ Sketch the 1-Nearest Neighbor class boundaries for this set of sample points. Clearly label the class on either side of the boundary.

PART B: Phoneme Recognition Experiments Utterances of 8 vowel phonemes from 38 speakers from the TIMIT database were extracted (about 2300 utterances total). Your goal for this problem is to achieve the highest possible recognition accuracy for this speech corpus. In the end you are free to do whatever you can to improve recognition accuracy. A demo Matlab program using 10 linear prediction coefficients and 1-nearest neighbor is provided to demonstrate usage of the database. The following files are provided: hw4Data.mat: Matlab .mat file containing the variables vocab, *Utter, and *Speaker, where * is one of the phonemes in vocab. *Utter is a 512xN (N varies w/ phoneme) matrix where each column is a 512-pt utterance extracted from the center (to minimize coarticulation effects) of a labeled phoneme from the TIMIT database. *Speaker is an Nx5 character matrix where each row i is the speaker label for column i in *Utter. *Speaker is provided to ensure that no speaker appears in both the test/training sets. The size is about 9MB uncompressed. hw4Demo.m: Matlab .m file that demonstrates usage of hw4Data.mat. LPC coeffs are extracted in bulk, random test/training speakers are designated for each classifier trial, and the test/training LPC coeffs are used w/ a 1-Nearest Neighbor classifer. Run this program to make sure you have downloaded the database properly. Tweak the following variables to see their effects on accuracy: percentTest, numTrials, vocab (you can shrink the vocab as a sanity check that your program works properly-small vocab means high recognition accuracy). Feel free to modify this code when writing your own solution. hw4Readme.txt: Readme file that describes all files in hw4.zip. All these files are conveniently available in hw4.zip which can be found at:
http://www.cnel.ufl.edu/hybrid/courses/EEL6586/hw4.zip

B1

Choose a robust feature extraction technique that you think will provide best results. You may use any feature extraction techniques you want (energy, zero-crossing, LPC, mfcc, PLP, hfcc...) or any combination of these. You are free to look at several different types of feature sets or to invent your own but do whatever you can to improve the recognition accuracy (without using test data during training). Explain your choice of feature set and why you think that it should perform well.

B2

Use a classification algorithm to classify the test data. Again you are free to use any classifier you like (Nearest Neighbor, Bayes, Neural Network, HMM, ...) Briefly explain why your choice of a classifier is a wise one (even if you decide to stay with the nearest neighbor classifier).

B3

Always include several trials as in hw4Demo.m and report the average over all trials. For your final version, make sure to include at least 100 trials. What is your final accuracy rate? What is the standard deviation of your accuracy value?

B4

For your final optimized system, which two phonemes are most likely to be confused with one another?

B5

Comment on why it is important that no speaker appear in both the test/training datasets.

As usual, attach all of your code to the end of the assignment. A total of 5 Bonus points will be awarded to the person(s) with the highest percentage correct classification.