Speech Processing by Computer

LECTURE 5

FORMANT AND FUNDAMENTAL FREQUENCY ANALYSIS

Objectives

By the end of the session you should:

• be able to describe two methods for separating information about source and filter in a spectrum

• understand the problems associated with estimating formant frequencies

• be able to describe one method for formant frequency estimation

• understand the problems associated with estimating fundamental frequency

• be able to describe one method for fundamental frequency estimation

Outline

5. Formant and Fundamental Frequency Analysis

5.1. Source-filter separation methods

5.1.1. Spectral modelling (LPC)

5.1.2. Cepstrum

5.2. Formant Frequency Analysis

5.2.1. General issues

5.2.2. From cepstrally-smoothed spectra

5.2.3. From LPC spectrum

5.3. Fundamental Frequency Analysis

5.3.1. General issues

5.3.2. Time-domain methods

5.3.2.1.Autocorrelation

5.3.3. Frequency-domain methods

5.3.3.1.Cepstrum

5.3.4. Combination methods

5.3.4.1.Inverse filtering

Reading

T. Parsons, Voice and Speech Processing, McGraw-Hill, 1987, Chapter 8.