1 Data collection techniques in the USF speech production and perception laboratory, and their reliability Wodzinski, S.M., Hardin, S.A., Frisch, S.A.,

1

Data collection techniques in the USF speech production and

perception laboratory,and their reliability

Wodzinski, S.M., Hardin, S.A.,

Frisch, S.A., & Stearns, A.M.,

University of South Florida

[email protected]

This work supported by NIH-NIDCD R03 06164

2

Ideal Recording Setup

• Head stabilized– Head location fixed– Probe location (relative to head) fixed– Stone (2004): Need consistent point of

reference in images for measurement (cf. Whalen 2004 HOCUS system)

3

Ideal Recording Setup

• Jaw able to move freely– If probe location is fixed, jaw contact with

probe makes this difficult– Need for ‘compressible acoustically

transparent standoff’ (Stone 2004)• Compressible so jaw can move

• Acoustically transparent so it does not interfere with ultrasound beam

4

USF Ultrasound Lab

5

Head Stabilization

• Head location fixed by a ‘halo’– Current version pieced together with rods

and clamps borrowed from the physics laboratory

– Similar in spirit to the HATS system (Stone & Davis 1995)

– FYI, body movement minimized by placing participant in a non-wheeled straight backed chair

6

Head stabilizing ‘halo’

7

Head Stabilization

• Participant rests forehead against headrest (padded dowel)

• Side and rear padded dowels adjusted to press lightly on the head

• This arrangement discourages head movement, but does not keep the head held absolutely fixed

8

Padded dowel

9

Demonstration of ‘halo’ on SAH

10

Compressible acoustically transparent standoff (CATS)

• Probe held by clamp on a cross-bar within the same apparatus– Raised to fit snugly under the chin– CATS between probe and chin allows jaw

movement without disturbing the image– Currently, CATS held on by rubber bands– Unfortunately, CATS does provide some

resistance to jaw movement

11

CATS

• CATS is a gel computer wrist rest– Trial and error to find a good wrist rest for

this purpose– Recommend something that comes inside a

cloth sleeve– Some rests within a non-transparent plastic

sleeve have a more gelatinous interior and do not hold shape without the sleeve (yuck!)

– Watch for air bubbles

12

CATS in USF lab

13

Other Info from Our Lab

• Using Aloka SSD-1000

• Record directly to computer with Canopus ADVC-1394

• Have a computer with two 17” screens for data analysis (not pictured)

• Second hard drive and DVD writer for storage and archiving

14

Measurement

• View video with Adobe Premiere

• All studies focusing on stops /k, g, t, d/

• Identify closure visually with reference to the waveform in Premiere

• Extract frame to Photoshop for measurement

15

Reliability of Measures

• Hardin (in progress) evaluating the reliability of hand measures of stop consonants

• Data presented today from a reliability study of velar stop closure measures in Wodzinski (2004)

16

Wodzinski (2004)

• Presented (in part) at Ultrafest II

• Study examined fronting of velar stops by three normal adult participants

• Found velar stop closure location (quantified by an angle measure) correlated strongly with the frontness of the following vowel (quantified by F2)

17

/gev/

/gu/

18

Reliability of Measures

• Raters– 2 female research assistants– Enrolled in the USF speech-language

pathology master’s degree program– Trained in phonetics and speech science,

and experienced in ultrasound measurement

19

Reliability of Measures

• Procedure– Raters used the same computer programs

and settings– Measurements done completely

independently (nearly a year apart)– Raters given same criteria for measurement

20

Reliability of Measures

• Criteria used by researchers to identify dorsum closure location:– Direction of tongue movement into and out

of closure– Tongue flattening against hard palate at

closure location– Brighter margin at edge of tongue surface

during closure (probably because that portion of the tongue is no longer in motion)

21

Reliability of Measures

• Measurement procedure– Closure is identified– The most anterior and posterior points of

closure are marked– A midpoint is determined between the

points by computer algorithm– The dorsal angle is derived between

midpoint and the center point of the probe at the bottom of the ultrasound image

22

Results (Closure frame)

• Raters compared for selection of video frame to measure as midpoint of velar closure

• Analysis of difference in frame number between repeat of measure (SH) and original measure (SW)

23

Results (Closure frame)

Frame diff Word data Nonwd data

-3 1% 0%

-2 3% 0%

-1 10% 9%

0 53% 76%

1 27% 16%

2 6% 0%

3 1% 0%

24

Results (Closure points)

• Raters compared on position of anterior and posterior closure points

• Word data– Anterior point avg 3.2 mm difference– Posterior point avg 6.2 mm difference

• Nonword data– Anterior point avg 3.1 mm difference– Posterior point avg 5.2 mm difference

25

Results (Dorsum angle)

• Resulting dorsum angle compared between raters

• Significant differences between dorsum angle measures between raters– Word data SH < SW by 1.6º [t(119) = 10.3,

p < 0.01]– Nonword data SH < SW by 0.7 º [t(171) =

4.1, p < 0.01]

26

Results (Dorsum angle)

• However, raters agreed in the overall pattern for how dorsum angle related to following vowel

• Correlation between raters– Word data, r = 0.93– Nonword data, r = 0.95

27

Results (Dorsum angle)

• Qualitatively, the resulting analysis of coarticulation between vowel and velar angle is the same for both raters

60708090100110120

iiheehaeuheracooouaiauoi

Vo

we

l

Angle (Max-Min-Mean)

60708090100110120

iiheehaeuheracooouaiauoi

Vo

we

l

Angle (Max-Min-Mean)

28

Summary (Reliability)

• There were quantitative differences between raters in their measures of velar closures– Inherently imprecise methodology?– Strategies of raters for identifying closure?– Experience with measuring ultrasound

images?

• Nonword data appeared more consistent than word data between raters… why?

29

Summary (Lab tech)

• The USF ultrasound lab has functional but probably not permanent solutions to the problems of head stabilization, probe stabilization, and a compressible standoff

• Though these solutions are temporary, we have collected informative data on the articulation of stop consonants

30

Question

• At USF, we have had to throw away about 50% of our participants because of ‘fuzzy’ tongue, or image drop out in the production of /k, g, t, d/– Is there a modification of apparatus that

would increase # of usable participants?– Is there a modification of collection

technique that would increase # of useable participants?