Paul Milenkovic

Acoustic characteristics of children's voice

Summary Four acoustic measures were taken from voice recordings of 121 children between the ages of 64 and 134 months. Acoustic parameters were measured on a sustained neutral /a/ vowel, produced imitatively at normal pitch and loudness. Samples were analyzed for fundamental frequency, jitter, shimmer, and signal-to-noise ratio (SNR). Data are presented to characterize effects of age, sex, height, and weight on these acoustic parameters. Results indicate statistically significant relationships between frequency and sex, with higher frequencies for girls. Also significant is a positive relationship between shimmer and height and a negative relationship between SNR and height.

Comparison of young adult singers and nonsingers with vocal nodules

Summary Comparisons were made between 10 singers and 10 nonsingers with vocal nodules and two control groups of normals, 10 singers and 10 nonsingers, on a wide range of acoustic, aerodynamic, psychoacoustic, and videostroboscopic measures. Results showed significant differences between the normals and those with nodules as well as differences between the singers and nonsingers. The singers with nodules had smaller nodules, less impairment of vibratory function, and less severe vocal symptoms than their nonsinging counterparts. The singing normals were found to be superior to the nonsinging normals on acoustic measures including jitter, shimmer, and signal-to-noise ratio. In addition, the singers, even in the presence of nodules, had superior maximum performance skills than their nonsinging counterparts.

Analyses of the hidden units of the multi-layer perceptron and its application in acoustic-to-articulatory mapping

An artificial neural network (ANN) is applied to perform the task of acoustic-to-articulatory inversion. The objective is to model the highly nonlinear mapping from linear predictive coding (LPC) code to corresponding articulatory parameters with a multilayer perceptron ANN structure. Such information will facilitate the study of the relationships between the acoustic signal and the physical vocal tract which produces it. Several novel approaches for devising the ANN structure have been evaluated. Specifically, the performance of two learning algorithms, a backpropagation (BP) algorithm, and a random optimization (RM) algorithms, are compared. To reduce excessive, redundant hidden units in the multilayer perceptron model, a singular value decomposition is applied to either the weight matrix or the output covariance matrix of the hidden units to check their corresponding ranks. In both cases, their ranks are closely related to the number of essential decision regions in the input data.<<ETX>>

Acoustic tube reconstruction from noncausal excitation

Algorithms and experimental procedures are presented for the reconstruction of the area function of a nonuniform tube. This reconstruction is performed from acoustics measurements that are taken in response to acoustic excitation of the tube at one end. The novel aspect of the procedure is that it is applicable to the case where the excitation waveform is a noncausal signal, which includes the case of a periodic excitation signal where the fundamental period is much shorter than the acoustic decay time of the tube being measured. Measurements are made of hard-walled pipes of known area as well as of the human vocal tract.

A fast least-square deconvolution algorithm for vocal tract cross section estimation

An efficient computation algorithm, based on unitary Jacobi-type rotations, is developed for fast least-square deconvolution. Specifically, this algorithm is able to solve a positive definite covariance matrix with a special Toeplitz structure in O(N/sup 2/) operations instead of O(N/sup 3/) operations. This algorithm, which takes advantage of the inherent structure of the underlying deconvolution problem, is guaranteed to be numerically stable since only unitary transformation is used. It is implemented on an experimental system for the estimation of human vocal tract cross-section. A significant gain in processing is observed. >

Series Solution for Finite Displacement of Planar Four-Bar Linkages

A time-varying instantaneous screw characterizes the motion of a rigid body. The kinematic differential equation expresses the path taken by any point on that rigid body in terms of this screw. Therefore, when a revolute joint is attached to a moving link in a planar kinematic chain, the path taken by the center of that revolute joint is the solution to such an equation. The instantaneous screw of a link in that chain is in turn determined by the action of the joints connecting that link to ground, where the contribution of each joint to that instantaneous screw is determined by its actuation rate and center point. Substituting power series expansions for joint rates into the kinematic differential equations for joint centers, and expressing loop closure as a linear constraint on the instantaneous screws of the links, a recurrence relation is established that solves for the coefficients in those power series. The resulting solution is applied to determine the equilibrium pendulum tilt of the United Aircraft TurboTrain. Comparing that power series approximation with an exact kinematic analysis shows convergence properties of the series.

Triangle Pseudocongruence in Constraint Singularity of Constant-Velocity Couplings

Congruent triangles establish that a class of intersecting-shaft couplings is constant velocity. These mechanisms employ a pair of linkages in parallel: a spherical joint at the intersection of the shafts and the intersection of straight-line tracks away from the shaft center to transmit rotation. A proof of constant velocity follows from the congruence of an initial pair of triangles with two matching sides and one excluded angle. This side-side-angle (SSA) condition is a pseudocongruence because it allows two different values for the included angle, indicating that such shaft couplings have symmetric and skewed assembly configurations. If the other excluded angle happens to be 90 deg, the SSA condition guarantees congruence because there is a single solution for the included angle. The 90 deg condition, however, occurs at a posture with a constraint singularity, where the shaft coupling is unable to transmit torque. Motion screw analysis establishes the same geometric condition for a coupling based on a revolute-spherical-revolute Clemens linkage. An upper bound on shaft deflection imposed by hyperextension of that linkage, along with a bound on deflection where constraint singularity occurs, identifies couplings where the constraint singularity can occur within the physical limits.

Effect of the vocal tract yielding sidewall on inverse filter analysis of the glottal waveform

Summary The inverse filter is a serial cascade of filter elements with a transfer function that cancels the effect of the poles of the vocal tract transfer function on the acoustic waveform to reveal the underlying glottal volume velocity waveform. Inaccuracies in the glottal wave reconstruction derived from an all-zero inverse filter can be attributed to deviations of the vocal tract transfer function from an all-pole model. Presented is an analysis of the error stemming from the effect of the yielding vocal tract sidewalls on the vocal tract transfer function. Predictions about the resulting artifacts in the estimated glottal volume velocity are derived from an acoustic model. These predictions are confirmed by applying a linear predictive coding (LPC) inverse filter analysis method to vowels synthesized using a transmission line model of the vocal tract containing yielding sidewall parameters as well as natural productions of nonnasalized vowels.

Nonsingular Spherically Constrained Clemens Linkage Wrist

The parallel connection of a revolute-spherical-revolute Clemens linkage with a revolute-universal joint (RU) spherical linkage realizes a two-axis spherical pointing device suitable for use in a robot wrist. The design is derived from a Clemens linkage constant-velocity coupling having a spherical constraint. A novel transverse orientation of the RU linkage facilitates control of the pointing direction through actuation of a pair of revolute joints at the base of the device, a capability not otherwise required when this type of mechanism is employed as a constant-velocity coupling. As the mechanism is confined to spherical space, spherical trigonometry and related geometric reasoning may be employed in the analysis of mobility and actuation. The analysis derives the forward and inverse kinematic relations, showing that the mechanism allows a full 180 deg of deflection of the pointer from the center before encountering a singularity.

Classification of vowels using spectrum moments

Formant frequencies form a set of acoustic features that can be used to classify vowels phonetically [A. K. Syrdal and H. S. Gopal, J. Acoust. Soc. Am. 79, 1086–1100 (1986)]. The advantage of formant frequencies is that they are unaffected by changes in the speech spectrum slope attributable to the voice source. The disadvantage of formant frequencies is in the large errors in frequency that can result from a misclassification of the formant peaks of the speech spectrum. A set of features based on statistical moments computed from the speech spectrum has been successful in classifying a broad class of obstruent sounds [K. Forrest et al., J. Acoust. Soc. Am. Suppl. 1 82, S84 (1987)]. In the case of vowel sounds, moment analysis eliminates the problem of identifying formant peaks but introduces the problem of variability in the moment features resulting from changes in the voice source. In order to correct for between subject variation in the voice source, the long‐term average spectrum of all of the speech...