Anatomic Aspects

• Larynx: Sytem of muscles, cartileges and ligaments.

• Primarly for controlling vocal folds• Folds 13-15mm

• Glottis: Slit like opening between folds• Folds move and change tension• Vocal folds and epiglottis close during

eating (also false vocal folds)• False vocal folds are likely open during

speech prouction

Larynx

Top View

Voicing Breathing

Three Primary States of Vocal Folds: Breathing, Voiced and Unvoiced

• BREATHING: Arytenoid cartileges move outward, fold muscles are relaxed.

• VOICING:The size/mass and tension of folds determines pitch (60-400 Hz)

MOTION OF FOLDS DURING VOICING

Mechanical Model of Folds

Periodic Glottal Flow Velocity

Mathematical Model of Glottal Flow

Glottal flow waveform

npngnu *

Glottal flow over a single cycle (imp. Resp.)

Impulse train

;

k

kPnnp

P : pitch period

Looking through a small window ,nw

npngnwnu *,,

Absence of window would yield impulses

k

k

GWP

U

*,1

,

Practical Phenomena

• Fixed pitch is not possible even in sustained cases “pitch jitter”

• Airflow velocity within a glottal cycle may differ across consecutive periods “shimmer”

• Jitter and shimmer yield “naturalness” and contributes to the “voice character”.

UNVOICING no vibration of vocal folds

Folds are close to each other but open and tense

Turbulance / aspiration/h/ whisper

• Aspiration normally occurs during voicing also.• If strong breathy voice.

Other Forms of Vocal Fold Movement

• Creaky Voice: Folds are very tense and vibrating part is small. – It is a harsh sounding voice.– High and irregular pitch is observed.

• Vocal Fry: When folds are massy and relaxed.– Abnormally low and irregular pitch.– Secondary glottal pulses arise within the open phase– Folds may couple with false vocal folds.– Vocal fry may ocur in normal speech at the edn of a phrase when

folds relax and lung pressure decreases.

• Diplophonic: Secondary pulses in the closed phase.– Generally in low pitch speakers.– It may arise in normal voices at the end of a phrase or word.

• Vocal fry and diplophonic can be modelled as

0~ nngngng

Vocal Fry

Diplophonic

Vocal Tract

~17 cm and up to ~20 cm2

• Vocal tract colors the source• It also generates new sources for sound production• It can be approximated by a linear filter.

– Resonance frequencies are called “formants”

– Formant bandwidth and formant amplitude.

• The peaks of the spectrum of vocal tract response are approximately formants.

00r

formant

00

jer

Related to bandwidth

All-pole model

• ck : pole, resonance freq., formant

• Formants: F1, F2, ... From low-to-high

• In general, formant frequencies decreaseas vocal tract length increases. (male, female, child)

iN

kkk zczc

AzH

1

1*1 11

Vocal Tract Output

Vocal tract impulse response

npngnhnx **

In a small window (to pick a stationary portion)

nxnwnx ,,

kkGHW

PX *,

1,

k

kkk WGHP

X ,1

,

Speech Sounds - Phonemes

• Broad classification is “vowels” and “consonants”.• Finer classification:

Vowels

Front

Center

Back

Consonants

Nasals

Plosives

Whispers

Fricatives

Voiced

Unvoiced

Voiced

Unvoiced

Transitional

Affricates

chew

just

Diphtongs

hide

boy

out

new

Semi-Vowels

Liquids

read

let

Glides

we

you

• Vowels: No constriction along the vocal tract.– Voiced. Vocal folds vibrate. – Front, center or back; according to the tongue position.– Rounded, unrounded according to the position of lips.– Also, high-low according to tongue height.

• Normal breathing yields low sound. The high intensity of vowels is because of vocal fold vibration.

Vowels

Fricatives

Plosives

Nasals

Transitional Sounds

Spectrogram of the diphtong /O/ in “boy”.