NEURAL NETWORK MODELING OF SPEECH PRODUCTION

Boston University, Boston MA

Investigators

Linked publications & trials

Abstract

The long-term objectives of the proposed research are to elucidate the stages of speaking skill development in infants and the motor control mechanisms for speech production in adults. These research areas are important for early diagnosis and proper treatment of speech disorders. The proposed research consists primarily of the development, refinement, and experimental testing of a comprehensive neural network modeling framework for speech production based on preliminary work described in Guenther (Appendices A). Model additions will include an articulatory mechanism that allows synthesis of speech wave-forms and an acoustic-like coordinate frame for speech movement planning. Proposed research also includes the development of software for producing speaker-specific vocal tract models based on Magnetic Resonance Imaging (MRI) scans. These models will allow synthesis of speech signals that account for the differences in vocal tract sizes and shapes of different individuals, thus providing a more accurate means for investigating the acoustic/articulatory relationships of individuals acting as subjects in speech production experiments. An experimental investigation utilizing these speaker- specific vocal tract models is also proposed to test an hypothesis generated by the modeling framework. Some speakers use two entirely different articulator configurations, called "bunched" and retroflex", to produce /r/ in different contexts. The proposed model predicts that the same target is specified tot he production mechanism in the two cases, but that different articulator configurations arise in different contexts due to two properties of the speech production mechanism: (1) movement planning in an acoustic-like coordinate frame, and (2) transformation of the planned acoustic trajectories into articulator movements via a direction-to- direction mapping. Speaker-specific vocal tract models corresponding to two subjects will be incorporated into the proposed modeling framework, which will then be used to predict which configuration each subject will use to produce /r/ in each of four contexts. The hypothesis will be tested by comparing model performance with the performance of the subjects while producing /r/ in the same four contests, as measured in an Electro-Magnetic Midsagittal Articulometer (EMMA) study. Finally, the proposed modeling framework will be used to investigate several other issues in speech production, including two modeling studies of speech motor development in infants (made possible by the self-organizing nature of the proposed model), and an investigation of intrinsic timing issues. GRANT=R01DE09161 The ability to utilize hemin and hemin containing compounds as an iron source has been documented for several pathogenic bacteria, including the periodontopathogen, Porphyromonas gingivalis. We have previously determined that P. gingivalis transports the entire hemin moiety into the cell by an energy-dependent mechanism and that the binding and accumulation of hemin are induced by growth of cultures in the presence of hemin. However, the specific P. gingivalis components involved in hemin binding and transport have not been identified. Growth of P. gingivalis under hemin-replete conditions has also been shown to influence the expression of several virulence factors; however, the role of hemin in the regulation of specific virulence genes has not been precisely defined. The primary objectives of the present application are to define the molecular mechanisms involved in hemin binding and transport in P. gingivalis and to examine the regulation of hemin responsive genes. Four specific aims are proposed: 1. To identify and characterize athe P. gingivalis hemin receptor(s). P. gingivalis outer membrane proteins involved in hemin binding will be identified by hemin affinity chromatography. The specificity of the putative receptor(s) will be defined by examining the binding of 14[C]hemin to P. gingivalis in the presence of hemin and nonahemin iron sources. 2. To clone P. gingivalis genes encoding proteins involved in hemin binding and transport. P. gingivalis genes encoding hemin binding proteins will be cloned by screening E. coli recombinants for the ability to bind hemin. P. gingivalis genes encoding proteins involved in hemin transport will be cloned by screening E. coli hemA mutants for the ability to grow with hemin. Corresponding P. gingivalis mutants will be obtained by insertional inactivation of the cloned genes and characterized both in vitro and in vivo. 3. To elucidate the regulation of hemin binding and transport genes. The regulation of cloned P. gingivalis genes involved in hemin binding and transport will be examined by analysis of mRNA and transcriptional fusions under hemin-deplete and -replete conditions. 4. To identify P. gingivalis genes that are regulated by the ferric uptake regulator (Fur). Our preliminary results indicate that the expression of hemin/iron-responsive genes in P. gingivalis may be controlled by the negative ferric uptake regulator protein, Fur. We will begin to identify P. gingivalis genes that are regulated by Fur by screening a P. gingivalis genomic library using the Fur titration asssay. The results obtained in these studies will allow us to identify specific components of the hemin transport system in P. gingivalis and will provide important information on the regulation of hemin responsive genes.