You are on page 1of 2

EE6641 Analysis and Synthesis of Audio Signals

HW#5: Linear Prediction Speech Processing


Due May 29 (Tue), 2018 by the end of night.
Prof. Yi-Wen Liu

This homework is aimed to familiarize you all with the ideas in linear prediction
(LP)-based speech analysis. Please record your own pronunciation of /a/(啊),
/i/(衣), /u/(烏) and the word 迂, save it as .wav files, and run the code
<ASAS_HW5_LinPred_2018.m> to process the files. Do the following
investigations.

1. Adjust the frame length and the order of LP (please look for the labels
[INVESTIGATE] throughout the comments in the code). Listen to the resulted
excitation signal excitat. Find out what range of these two parameters (p and
framelen) results in successful removal of the original vowel quality.

2. (Optional, for those of you that are theoretic-inclined) Study how LP


coefficients are calculated in the code, and try implementing the
Levinson-Durbin algorithm by yourself. Check whether it is the same as given by
Matlab’s levinson() function.

3. Investigate how stationary the estimated K parameters (stored in the array


Kcoeff) are as they inevitably vary across frames. Since the parameters are
intimately related to the shape of the vocal tract, do they remain more consistent
if you try to sustain your vowel quality by holding your articulators still, such as
your jaw position and your tongue position? How about your pitch? Does it help
to keep Kcoeff stationary by keeping your voice at the same pitch (instead of
singing the vowels or making any inflection)

4. Write your own code to estimate the frequency of the first three formants of
each of the vowels. Perform pairwise comparison between:
Ø /a/ and /i/
Ø /i/ and /u/
Ø 衣 and 迂
Formatting and grading policy
This homework will be graded based on your written report. Please do not
exceed 8 pages. Please do not use a fontsize greater than 12. You are more than
welcome to compare your results with other people’s pronunciation and discuss
about the inter-personal differences in the report.

When you submit this homework, please include a .pdf file of the report, and put
everything else in a .zip file.

Blessings!

-------
ps. The last part of the demo code is my attempt to convert Kcoeff into the
shape function of the vocal tract.

You might also like