Computer Science and Engineering
Permanent URI for this collection
Browse
Browsing Computer Science and Engineering by Author "Baljko, Melanie"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Open Access Consonant Classification Based on Tongue Tip Trajectories(2018-08-27) Zarringhalam, Rojin Majd; Baljko, MelanieIn this thesis, I investigate an issue that is foundational to the development of a new class of novel game-based speech therapies. Whereas several prior computer-based approaches have focused on the use of clinical objectives that concern spatialized aspects of the tongue-tip trajectory (e.g., the targeting of improved accuracy in lingual-palate contact for certain phonemic segments), this line of inquiry focuses on the potential use of attributes relating to the speed and velocity of the tongue-tip trajectory as an alternative clinical objective. I situate my work in the body of prior work on the velocity characteristics of different phonemic segments. For speed and velocity-based clinical targets to be viable, however, it is necessary to characterize and to analyze the relative amounts of variability among and within talkers and phonemic segments with respect to speed-related characteristics. I describe our approach and the results of an analysis which focuses on a large kinematic speech dataset that includes multiple repetitions of 8 different phonemic segments (/t/, /d/, /k/, /g/) as plosives, (/s/, /sh/, /z/) as fricatives and (/tch/) as affricate by 17 talkers. Last, we provide an illustration of how such normative (albeit speaker-dependent) speed and velocity profiles would be instantiated via an interactive scenario that could be included within an extant computer-based speech therapy system. I will represent the classification accuracy results of kinematic data using HMM and SVM classification techniques.Item Open Access Tracking Visible Features of Speech for Computer-Based Speech Therapy for Childhood Apraxia of Speech(2018-03-01) Zhian, Zhian; Baljko, MelanieAt present, there are few, if any, effective computer-based speech therapy systems (CBSTs) that support the at-home component for clinical interventions for Childhood Apraxia of Speech (CAS). PROMPT, an established speech therapy intervention for CAS, has the potential to be supported via a CBST, which could increase engagement and provide valuable feedback to the child. However, the necessary computational techniques have not yet been developed and evaluated. In this thesis, I will describe the development of some of the key underlying computational components that are required for the development of such a system. These components concern camera-based tracking of visible features of speech which concern jaw kinematics. These components would also be necessary for the serious game that we have envisioned.