Using statistical decision theory to predict speech intelligibility. I. Model structure.

This article introduces a new model that predicts speech intelligibility based on statistical decision theory. This model, which we call the speech recognition sensitivity (SRS) model, aims to predict speech-recognition performance from the long-term average speech spectrum, the masking excitation in the listener's ear, the linguistic entropy of the speech material, and the number of response alternatives available to the listener. A major difference between the SRS model and other models with similar aims, such as the articulation index, is this model's ability to account for synergetic and redundant interactions among spectral bands of speech. In the SRS model, linguistic entropy affects intelligibility by modifying the listener's identification sensitivity to the speech. The effect of the number of response alternatives on the test score is a direct consequence of the model structure. The SRS model also appears to predict the differential effect of linguistic entropy on filter condition and the interaction between linguistic entropy, signal-to-noise ratio, and language proficiency.     

Tumor-associated antigens identified by mRNA expression profiling as tumor rejection epitopes

Thirteen H-2b-binding peptides derived from six potentially overexpressed proteins in p53-/- thymoma (SM7) cells were studied for immunogenecity and vaccine-induced prevention of tumor growth in mice inoculated with SM7 tumor cells. Six of the peptides generated specific CTL responses after immunization, but only two of these peptides (RAD23-31 and RAD24-31) were capable of generating a weak vaccination-induced protection against adoptive tumor growth. SM7 inoculated mice treated with a blocking antibody against the inhibitory T cell signal transducing molecule CTLA4 appeared to delay tumor take, suggesting that SM7 thymoma cells are recognized by the adaptive immune system of the host. However, prophylactic vaccination with RAD23-31 and RAD24-31 peptides combined with anti-CTLA4 Ab treatment and did not improve tumor resistance. Our data would indicate that vaccination with immunogenic peptides derived from potentially overexpressed tumor proteins, as identified by mRNA expression profiling of p53-/- thymoma cells, at best results in a weak tumor protection thus questioning this way of detection of new tumor rejection epitopes.     

Interactions between test- and inducer-tone durations in induced loudness reduction

A tone usually declines in loudness when preceded by a more intense inducer tone. This phenomenon is called "loudness recalibration" or "induced loudness reduction" (ILR). The present study investigates how ILR depends on level, loudness, and duration. A 2AFC procedure was used to obtain loudness matches between 2500-Hz comparison tones and 500-Hz test tones at 60 and 70 dB SPL, presented with and without preceding 500-Hz inducer tones. For 200-ms test and comparison tones, the amount of ILR did not depend on inducer level (set at 80 dB SPL and above), but ILR was greater with 200- than with 5-ms inducers, even when both were equally loud. For 5-ms tones, ILR was as great with 5- as with 200-ms inducers and about as great as when test and inducer tones both lasted 200 ms. These results suggest that (1) neither the loudness nor the SPL of the inducer alone governs ILR, and (2) inducer duration must equal or exceed test-tone duration to yield maximal amounts of ILR. Further analysis indicates that the efferent system may be partly responsible for ILR of 200-ms test tones, but is unlikely to account for ILR of 5-ms tones.     

Solving multistage stochastic network programs on massively parallel computers

Multi-stage stochastic programs are typically extremely large, and can be prohibitively expensive to solve on the computer. In this paper we develop an algorithm for multistage programs that integrates the primal-dual row-action framework with proximal minimization. The algorithm exploits the structure of stochastic programs with network recourse, using a suitable problem formulation based on split variables, to decompose the solution into a large number of simple operations. It is therefore possible to use massively parallel computers to solve large instances of these problems. The algorithm is implemented on a Connection Machine CM-2 with up to 32K processors. We solve stochastic programs from an application from the insurance industry, as well as random problems, with up to 9 stages, and with up to 16392 scenarios, where the deterministic equivalent programs have a half million constraints and 1.3 million variables. Research partially supported by NSF grants CCR-9104042 and SES-91-00216, and AFOSR grant 91-0168. Computing resources were made available by AHPCRC at the University of Minnesota, and by NPAC at Syracuse University, New York.     

Psychometric functions for gap detection in a yes-no procedure

To examine models of temporal resolution and to investigate the decision processes underlying the detection of a brief pause in a bandpass noise, psychometric functions for gap detection were measured at octave frequencies from 0.25 to 8 kHz. Three normal listeners were tested using a constant-stimulus procedure with a cued Yes-No paradigm. The Minimum Detectable Gap (MDG) estimated from the midpoint of the psychometric functions decreased systematically with increasing frequency. The slopes of the psychometric functions generally increased as the test frequency increased up to 2 kHz, but remained constant at the higher frequencies. Two models were investigated: an energy-detector model and a loudness-detector model. Both consisted of auditory filtering, a nonlinearity, and short-term integration. In the energy-detector model, the nonlinearity was a square law. In the loudness-detector model, it was a compressive power law. Using the usual Gaussian approximations, the energy-detector model fails at low frequencies because the probability distributions of short-term energy differ from Gaussian distributions. The probability distributions of short-term loudness closely follow Gaussian distributions. The loudness-detector model predicts the frequency dependence of the MDG quite accurately, except at 0.25 kHz. It also predicts psychometric functions that resemble the data at low frequencies, but the predicted slopes increase much less with frequency than the measured slopes. This result may indicate that the onset response to the trailing marker of the gap provides an important cue for detection of gaps with durations exceeding the MDG.     

Photoelectron spectroscopy of hydrated hexafluorobenzene anions

We present a synergetic experimental/theoretical study of hydrated hexafluorobenzene anions. Experimentally, we measured the anion photoelectron spectra of the anions, C6F6(-)(H2O)n (n=0-2). The spectra show broad peaks, which shift to successively higher electron binding energies with the addition of each water molecule to the hexafluorobenzene anion. Complementing these results, we also conducted density functional calculations which link adiabatic electron affinities to the optimized geometric structures of the negatively charged species and their neutral counterparts. Neutral hexafluorobenzene-water complexes are not thought to be hydrogen bonded. In the case of C6F6(-)(H2O)1, however, its water molecule was found to lie in the plane of the hexafluorobenzene anion, bound by two O-H...F ionic hydrogen bonds. Whereas in the case of C6F6(-)(H2O)2, both water molecules also lie in the plane of and are hydrogen bonded to the hexafluorobenzene anion but on opposite ends. This study and that of Schneider et al. [J. Chem. Phys. 127, 114311 (2007), preceding paper] are in agreement regarding the geometry of C6F6(-)(H2O)1.