Identification of Stress Placement in Speakers with and without Dysarthria Pamela Campellone Thomas DiCicco Rupal Patel.

Identification of Stress Placement in Speakers

with and without Dysarthria

Pamela Campellone

Thomas DiCicco

Rupal Patel

Background Traditional research focus in dysarthria

due to CP: Articulation More recent research: Prosody Acoustic findings: Preserved prosodic

control at vowel & word level

Background Are acoustic signals consistent and

reliable? Can humans and/or machines make use

of these signals? If prosody is a strength, can it be

harnessed to improve segmental clarity?

Research Questions Can listeners identify stress within phrases

produced by speakers with dysarthria & age-gender matched healthy controls?

How accurate is machine classification of prosodic contrasts?

Method Spoken database: 12 speakers with

dysarthria (DYS) & 12 healthy controls (HC)5 phrases (4 monosyllabic words) produced

with stress on 1 of the 4 words or neutrally 48 monolingual speakers of American

English served as listeners4 listeners per DYS-HC speaker pair

Listener Interface

86

88

90

92

94

96

98

100

1 2 3 4 5

Stress Location

% A

ccur

acy

A B C D N

HC Ave

DYS Ave

Acoustic Predictors of Listener Accuracy (p<0.01)

Machine Classification HC & DYS words classified as stressed

vs. unstressedHC accuracy: 98.1%DYS accuracy: 97.4%

Separate combinations of duration, intensity, & F0 used to determine which were most predictive

Accuracy by Acoustic Predictors

 Predictor HC DYS

Dur 81.7 77.8

Int 89.9 88.3

F0 96.7 96.2

Dur + Int 95.3 92.0

Dur + F0 96.6 96.5

Int + F0 97.1 96.6

Dur + Int + F0 98.8 97.4

Conclusions Unfamiliar listeners & machine classifier

both highly accurate Communicative potential of prosody

Clinically: scaffolding for improved intelligibilityApplication: communication aids which utilize

prosodic variation

Future Directions Examine productions of speakers with

varying etiologies of DYSDifferences in acquired vs. congenital?

Assess prosodic control in more varied speech tasks

Design comprehensive interventions incorporating speaker and listener variables