Marisha Speights 123 Joel MacAuslan 2 Noah Silbert 1 amp Suzanne Boyce 12 1 University of Cincinnati 2 Speech Technology and Applied Research 3 Auburn University ID: 785076
Download The PPT/PDF document "Automated Analysis of Syllable Complexit..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Automated Analysis of Syllable Complexity as an Indicator of Speech Disorder
Marisha Speights
1,2,3 ,Joel MacAuslan 2,Noah Silbert1, & Suzanne Boyce 1,2 1University of Cincinnati, 2Speech Technology and Applied Research, 3Auburn University
PurposeThis study was designed to examine the feasibility of the Syllabic Cluster algorithm in the SpeechMark® MATLAB toolbox as an automated approach for identifying differences in speakers with and without Speech Sound Disorders(SSD).
BackgroundIn the course of normal development, children master voluntary coordination of the motoric movements necessary for the utterance of complex syllables. 1-3 Development of well-formed syllables has been shown to be a significant predictor of later communication skills. 4-6Children with delayed speech production show atypical trends in the mastery of well-formed syllables, especially in continuous speech. 6-10
REFERENCESFogel, A., & Thelen, E. (1987). Development of early expressive and communicative Journal of the Acoustical Society of America, 111(4), 1action: Reinterpreting the evidence from a dynamic systems perspective. Developmental Psychology, 23(6), 747.Rvachew, S., & Bernhardt, B. M. (2010). Clinical implications of dynamic systems theory for phonological development. American Journal of Speech-Language Pathology, 19(1), 34–50.Smith, L. B., & Thelen, E. (2003). Development as a dynamic system. Trends in Cognitive Sciences, 7(8), 343–348.Oller, D. K, Eilers, R. E, Neal, A. R., & Schwartz, H. K. (1999). Precursors to speech in infancy: the prediction of speech and language disorders. Journal of Communication Disorders, 32(4), 223–245Oller, D. K., Niyogi, P., Gray, S., Richards, J. A., Gilkerson, J., J., Xu, D., … Warren, S. F. (2010). Automated vocal analysis of naturalistic recordings from children with autism, language delay, and typical development. Proceedings of the National Academy of Sciences, 107(3), 13354–13359.Oller, D. K. (2000). The emergence of the speech capacity. Psychology Press.Pharr, A. B., Ratner, N. B, & Rescorla, L. (2000). Syllable structure development of toddlers with expressive specific language impairment. Applied Psycholinguistics, 21(04), 429–449. Fell, H. J, MacAuslan, J., Ferrier, L. J., Worst, S., & Chenausky, K. (2002). Vocalization Age as a Clinical Tool. In Proceeding from ICSLP. Flipsen Jr, P. (2006b). Syllables per word in typical and delayed speech acquisition. Clinical linguistics & phonetics. Clinical Linguistics & Phonetics, 20(4), 293–301, Paul & Jennings, 1992, Pharr, A. B., Ratner, N. B, & Rescorla, L. (2000). Syllable structure development of toddlers with expressive specific language impairment. Applied Psycholinguistics, 21(04), 429–449.Hodson, B. W., Scherz, J. A., & Strattman, K. H. (2002). Evaluating communicative abilities of a highly unintelligible preschooler. American Journal of Speech-Language Pathology, 11(3), 236-242. (Kent, R. D. (1996). Hearing and believing some limits to the auditory-perceptual assessment of speech and voice disorders. American Journal of Speech-Language Pathology, 5(3), 7–23.Secord, W., & Donohue, J. S. (2014). CAAP: Clinical Assessment of Articulation and Phonology. Super Duper Publications.) Wiig, E. H., Secord, W., & Semel, E. M. (2004). CELF preschool 2: clinical evaluation of language fundamentals preschool. Pearson/PsychCorp.Martin, B., & Carle, E. (1984). Brown bear, brown bear. Puffin books. Young, E. C. (1991). An analysis of young children's ability to produce multisyllabic English nouns. Clinical Linguistics & Phonetics, 5(4), 297-316.Anderson, C., & Cohen, W. (2012). Measuring word complexity in speech screening: single‐word sampling to identify phonological delay/disorder in preschool children. International Journal of Language & Communication Disorders, 47(5), 534-541Stoel-Gammon, C. (2010). The Word Complexity Measure: Description and application to developmental phonology and disorders. Clinical linguistics & phonetics, 24(4-5), 271-282.
ResultsMultinomial logistic regression models were used to determine if the LM/Syll and Syll/Utts parameters predict speaker group.Two models were fit, one with adults as the reference category and one with typical child speakers as the reference category. Speech complexity predictors tested in the model: Landmarks per Syllabic Cluster (Lm/Syll) and Syllabic Cluster per Utterance (Syll/Utt).When examining the relationship of adult speakers to child speaker groups all parameters are significant in the model (p < .001). The WCM and LM/SC are significant predictors of disordered status in children when typical speakers are the baseline in the model at p <.001 and p = .05, respectively.LM/SC was a significant predictor of disordered status when complex words were measured alone (p <.001).Mixed-effects logistic regression models were used to analyze how well LM/SC and SC/Utt predict disordered group status in connected speech samples. Mixed models are useful for analyzing multiple observations within subjects, providing flexibility in modeling expected values, as well as between-subject differences simultaneously (Gelman & Hill, 2006).SC/Utt was a significant predictor when continuous speech samples were measured (p <.001).
MethodsParticipants10 Adults, 27 Children without SSD, 10 with SSD. Children were age 3-5.Recording Method: Shure wireless microphone on child vest. Stimuli Recorded from Administration of:The Clinical Assessment of Articulation and Phonology 2nd edition: CAAP-2.13The Clinical Evaluation of Language Fundamental-preschool-2nd Edition: CELF-P.14Child Story book repetition Brown bear, brown bear what do you see?15Typical/Disordered cutoff: Standard scores ≥ 80 on the CAAP-2 and CELF-P = typically developing.
Methods, cont.Child speech corpus includes roughly 76 words (46 words from CAAP and 30 from two published lists16,17) and 33 sentences from a child story book.Transcribed samples were scored and ranked using the Word Complexity Measure (WCM)18.Number of Landmarks per Syllabic Cluster and Syllabic Clusters per Utterance were measured.
DiscussionStrengths:Demonstrated feasibility of using a Automatic Syllabic Cluster Analysis for detecting group differences in speakers. The approach allows the analysis of words and sentences without the need for transcription. Limitations: Further research is warranted.to determine the approach as a potential diagnostic tool.Conclusion:Automated detection using The SpeechMark® Syllabic Cluster algorithm was found to be a feasible approach for identifying group differences
ACKNOWLEDGMENTS
This work was funded by United States National Institutes of Health (NIH)grants R43 and R44 DC010104, R42 AG033523, and R42 HD34686 to S.T.A.R. Corp. For more information and SpeechMark downloads (currently freeware in beta version): www.speechmrk.com
Word
WCM Word WCMRefrigeratorTrousersFingernail Scissors Glasses Grasshopper Glove Basketball Giraffe Stairs 10 87 7 7 7 6 6 6 6 Legs FlagSchool Bridge Treasure Elephant Helicopter Lollipop Hippopotamus Sleeping5 55 5 5 5 5 5 5 5
Table 1: High complexity words were selected by scoring 76 words in the data set using the WCM. Words scoring ≥ 5 were included in the final analysis.
Table 2. 76 Words: Significance ***p<.001, **p=.05. Multinomial logistic regression model fits for landmarks per syllable (top row), syllable clusters per utterance (middle row), and Word Complexity Measure (bottom row). Slope estimates (first column), standard errors (second column), z values (third column), and significance (fourth column).
Clinical Problem
Accurately describing the skills of children with severe motor-speech disorders is notoriously challenging, especially for connected
speech.
Transcription and analysis of connected speech samples require a larger time commitment and expertise in phonetic transcription.11This is a particular problem in studies of children with speech disorders that affect intelligibility. 12
Syllabic Cluster Analysis The SpeechMark ® acoustic landmark analysis system has been developed to automate the detection of abrupt acoustic events. Listeners use abrupt changes in the acoustic signal to make decisions about what was uttered 16The sequence and grouping of landmarks is related to how the speech was spoken. If spoken more canonically, as a string of intended syllables (dictionary form), more landmarks will be detected.
Research Questions
RQ 1:Does the Landmark per Syllabic Cluster parameter predict speaker group ?RQ 2: Does the Syllabic Clusters per Utterance parameter predict speaker group?
Table 3
: Logistic regression model fits for landmarks per syllable (top row) and syllable clusters per utterance (bottom row). Slope estimates (first column), standard errors (second column), z values (third column), and p values (fourth column). The slope estimate and standard error for LM/SC indicates that LM/SC is a very poor predictor of disordered group status. On the other hand, the slope estimate and standard error for SC/Utts indicates that SC/Utts is a useful predictor of disordered group status. The slope estimate of -0.21 indicates that for each unit increase in one SC/Utts, the probability of being in the TD group decreases by approximately 0.05.
Figure 1. Syllable clusters per utterance (SC/Utts) and group status. The x-axis indicates group status (TD = Typically Developing; D = Disordered), and the y-axis indicates SC/Utts. The boxplots indicate the median (thick horizontal line) and interquartile region (lower and upper box limits), with the notch indicating an approximate 95% confidence interval for the median. The whiskers indicate 1.5 times the interquartile range, with dots indicating data outside this range.
Two speakers one with and without SSD.
The LMs are placed at points when abrupt change of energy is occurring simultaneously across multiple frequency ranges at multiple time scales. Waveform with smoothed amplitude envelope and landmarks generated by
SpeechMark
® MATLAB Toolbox. Solid red line shows the interval of voicing. The dashed blue line indicates the grouping of landmarks into a syllabic cluster as per SpeechMark® MATLAB Toolbox. The dashed magenta line shows the grouping for the utterance.
EstimateSEz valuePr(>|z|)LM/SC.01.05.17.85SC/Utt-.21.031-6.85<.001
Typical Child
Predictor
B
SE B
z value
Sig.
Adult
Lm/SC
.
13
.
02
5.65
***
SC/
Utt
.
63
.
04
26.59
***
WCM
.
11
.
03
4.46
***
Disordered
Lm/SC
-.
07
.
03
-2.14
**
SC/Utt
.
02
.
05
.
81
WCM
-.
31
.
04
-10.02
***