Abstract

The purpose of this thesis is to improve the quality of synthetic to attempt to speech by using the bes t excitation signal during Linear Predictive Coding (LPC) synthesis. This thesis examines the human speech system as a basis for our synthetic speech model. Then it closely examines LPC synthesis, including the mathematical details. One dominant factor in producing natural-sounding and intelligible speech is the excitation signal. For LPC synthesis the excitation signal must have a flat frequency spectrum. A train of impulses separated by the pitch period of the speech has been the standard excitation signal for voiced speech in LPC synthesis. Unfortunately, speech produced using this excitation signal has an unnatural nbuzz". For natural-sounding speech, the excitation signal should resemble the glottal volume velocity waveform. The glottal volume velocity waveform is a measure of the excitation that produces natural speech and it does not have a flat frequency spectrum. This raises the question: what type of excitation signal should be used to produce the most natural-sounding speech possible? To answer this question, we examined six excitation signals that are currently being used in LPC synthesis. We also developed many new excitation signals to be used specifically for synthesizing natural-sounding speech. We experimented with the LPC parameters and these excitation signals to determine the conditions that produced the best speech. Then we compared five of the excitation signals in forced pair trials. We found that our new excitation, LF Impulse excitation, produced speech superior in overall quality (that is naturalness and intelligibility) to the others. We conclude, therefore, that LF Impulse excitation, or an excitation similar to it, should be considered when attempting to produce speech that is both natural-sounding and intelligible with LPC synthesis.

Notes

If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu

Graduation Date

1987

Semester

Fall

Advisor

Alsaka, Yacoub

Degree

Master of Science (M.S.)

College

College of Engineering

Format

PDF

Pages

136 p.

Language

English

Rights

Public Domain

Length of Campus-only Access

None

Access Status

Masters Thesis (Open Access)

Location

Orlando (Main) Campus

Identifier

DP0021488

Accessibility Status

Searchable text

Included in

Engineering Commons

Share

COinS