Reiner Wilhelms‐Tricarico

Speech motor control: acoustic goals, saturation effects, auditory feedback and internal models

Abstract A theoretical overview and supporting data are presented about the control of the segmental component of speech production. Findings of “motor-equivalent” trading relations between the contributions of two constrictions to the same acoustic transfer function provide preliminary support for the idea that segmental control is based on acoustic or auditory-perceptual goals. The goals are determined partly by non-linear, quantal relations (called “saturation effects”) between motor commands and articulatory movements and between articulation and sound. Since processing times would be too long to allow the use of auditory feedback for closed-loop error correction in achieving acoustic goals, the control mechanism must use a robust “internal model” of the relation between articulation and the sound output that is learned during speech acquisition. Studies of the speech of cochlear implant and bilateral acoustic neuroma patients provide evidence supporting two roles for auditory feedback in adults: maintenance of the internal model, and monitoring the acoustic environment to help assure intelligibility by guiding relatively rapid adjustments in “postural” parameters underlying average sound level, speaking rate and the amount of prosodically-based inflection of F0 and SPL.

Tongue surface displacement during bilabial stops

The goals of this study were to characterize tongue surface displacement during production of bilabial stops and to refine current estimates of vocal-tract wall impedance using direct measurements of displacement in the vocal tract during closure. In addition, evidence was obtained to test the competing claims of passive and active enlargement of the vocal tract during voicing. Tongue displacement was measured and tongue compliance was estimated in four subjects during production of /aba/ and /apa/. All subjects showed more tongue displacement during /aba/ than during /apa/, even though peak intraoral pressure is lower for /aba/. In consequence, compliance estimates were much higher for /aba/, ranging from 5.1 to 8.5 x 10(-5) cm3/dyn. Compliance values for /apa/ ranged from 0.8 to 2.3 x 10(-5) cm3/dyn for the tongue body, and 0.52 x 10(-5) for the single tongue tip point that was measured. From combined analyses of tongue displacement and intraoral pressure waveforms for one subject, it was concluded that smaller tongue displacements for /p/ than for /b/ may be due to active stiffening of the tongue during /p/, or to intentional relaxation of tongue muscles during /b/ (in conjunction with active tongue displacement during /b/).

An educational articulatory synthesizer, EASY

An articulatory synthesizer has been written for general educational and research use. It has been named Educational Articulatory Synthesizer, EASY, and written in the MATLAB programming language. In its current instantiation, the synthesizer performs calculations in the frequency domain and includes the effects of wall‐vibration, viscous and thermal boundary layer loss, and radiation loss. Sources anywhere from the glottis to the mouth can be specified. Further, the code allows for side branches other than the nasal tract so that sounds such as laterals can be specified. A low‐frequency aerodynamic module has also been included. Midsagittal shape is controlled by a hierarchy of flesh points. Points high in the hierarchy specify the overall shape by determining reference positions for points lower in the hierarchy, which, in turn, determine the shape locally. [Work supported by grant NIDCD‐001247 to CReSS LLC.]

Rational approximations of viscous losses in vocal tract acoustic modeling

The modeling of viscous losses in acoustic wave transmission through tubes by a boundary layer approximation is valid if the thickness of the boundary layer is small compared to the hydraulic radius. A method was found to describe the viscous losses that extends the frequency range of the model to very low frequencies and very thin tubes. For higher frequencies, this method includes asymptotically the spectral effects of the boundary layer approximation. The method provides a simplification for the rational approximation of the spectral effects of viscous losses.

Biomechanical and Physiologically Based Speech Modeling

Improvements in speech synthesis may be achieved through an increased understanding of the actual physiology and control of speech production. Toward this end, a three-dimensional dynamic finite-element tongue model is described, which is the first component of a research project directed at a physiologically based computer simulation of speech production.

A new 3D dynamical biomechanical tongue model

A new dynamical biomechanical tongue model is being developed to study speech motor control. In spite of its computational complexity, a 3D representation was chosen in order to account for various contacts between tongue and external structures such as teeth, palate, and vocal tract walls. A fair representation of tongue muscle anatomy is provided, by designing the finite element mesh from the visible human data set (female subject). Model geometry was then matched to a human speaker, so that simulations can be quantitatively compared to experimental MRI data. A set of 11 muscles is modeled, whose role in speech gestures is well established. Each muscle is defined by a set of elements whose elastic properties change with muscle activation. Muscles forces are applied to the tongue model via macrofibers defined within the mesh by muscle specific sets of nodes. These forces are currently specified as step functions. Boundary conditions are set using zero‐displacement nodes simulating attachments of tongue o...

Digital filters for accurate simulation of wave propagation losses in tubes

A boundary layer approximation for viscous damping in one‐dimensional wave transmission in a tube results in an irrational frequency‐dependent damping filter for sound propagation. This filter can be approximated with high accuracy by a rational filter function that can be obtained from Pade approximations or continued fraction expansion. Taking into account the viscous losses in a Kelly–Lochbaum structure that represents sound propagation in a tube with spatially varying cross section results in replacing the delay elements of the lattice filter for the loss‐free case by special recursive filters. The design, implementation, and applications of the filter structures will be presented.

Padé approximations for boundary‐layer losses in articulatory synthesis

To build an articulatory synthesizer it is necessary to model acoustic propagation in tubes with variable area. Acoustic propagation entails viscous and thermal losses, which are strongest at the vocal tract walls. In the standard boundary‐layer approximation irrational frequency laws containing the square root of frequency best represent these losses. Our immediate goal is, given a sound source distribution in the vocal tract, to efficiently calculate the sound output from the mouth. In order to use digital filter theory in this process, a rational approximation to the square root of frequency is sought in the form of Pade approximations. One implementation is obtained by modification of the Kelly–Lochbaum algorithm for calculating wave propagation in a tube, using a high over‐sampling rate. However, frequency‐dependent loss means that both the reflection coefficients and the time delays through a tube section are non‐constant functions of frequency—an assumption used in the Kelly–Lochbaum algorithm. The...

Continuum mechanical model of the tongue and mouth floor

In this new model, the mouth floor and the tongue body are treated as blocks represented by tri‐quadratic finite elements in which the muscles are modeled as continuous directed fields of the fiber direction. Collision with the hard palate and other rigid structures is modeled by subdividing the surface into small triangles for rapid collision detection and through calculating forces by a penalty method to enforce impenetrability of the rigid structures. A similar method is provided for the more complicated case of the collision between tongue body and soft floor of the mouth. The model can be coupled with a model of the jaw and hyoid where muscles may be modeled as contracting strings. The moving hyoid and jaw provide kinematic constraints for the tongue/mouth‐floor model. Active muscle stress is generated by a Hill model of muscle shortening combined with a rational extension for muscle elongation, and a polynomial curve represents the relation between muscle length and stress. The stiffness matrix and ...

A finite‐element template for tongue modeling

This report describes the design, data structure, and applicability of a refined finite‐element template for the human tongue and connected oral structures. The finite‐element model relies mainly on data sets from the Visible Human Project of the National Library of Medicine. The model is composed of macroblocks, which represent geometric subsections of the tongue. These blocks represent, in some cases, either individually or in combination, functional subsections, such as individual muscles. For each macroblock (or geometric region) a finite‐element mesh can be generated such that the whole of the tongue can be modeled by a mesh of finite‐elements, since the subdivision is compatible across block boundaries. The finite element template contains information about muscle tissue distribution of the tongue and velopharynx. Approximate muscle fiber directions are represented as direction fields. The model can be adapted to individual morphology if a set of morphological landmarks can be specified from measure...