This paper presents an extensive statistical analysis of the acoustic startle response of two vocal parameters: fundamental frequency (F0) and root-mean-square energy (E), as well as of the orbicularis oculi (eyeblink) surface electromyography (sEMG). An experiment was conducted in which fourteen participants were exposed to acoustic startle stimuli of varying parameters, i.e., intensity level, duration, rise time, and spectral type, during periods of sustained phonation. Voice recordings of the phonations were taken alongside several physiological signals, of which only the sEMG was analyzed in this paper. Response features (peak value, peak time, latency, rise time, fall time, and duration) were extracted on F0, E and sEMG data, and statistical analysis was conducted using linear mixed effects models to show the response behavior with the varying stimuli. The results for vocal F0 and E data were congruent with sEMG data and earlier work in the field. The results demonstrated that vocal analysis can be used as a feasible alternative to the sEMG eyeblink analysis of acoustic startle responses.

The paper presents a new method for all-pole model estimation based on minimization of the weighted mean square error in the sampled spectral domain. Due to discrete nature of the proposed distance measure, emphasis can be put on an arbitrary set of spectral samples what can greatly improve the model accuracy for periodic signals. Weighting can also be applied to improve the fitting in certain spectral regions according to any desired fidelity criterion. Iterative algorithm for determination of the optimal model is proposed and an exceptionally fast convergence rate is demonstrated. Accuracy of the estimation algorithm is verified on an example of a synthetic vowel for a broad range of pitch frequencies.

A new distance measure for all-pole modeling of voiced speech is introduced in this paper. It can easily be integrated within the concept of discrete Weighted Mean Square Error (WMSE) all-pole modeling, by a suitable choice of the modeling weights. The proposed weighting will address the problems such as: harmonic estimation reliability, perceptual significance of the harmonic and the model mismatch errors. The robust estimator is proposed, to reduce the effect of outliers caused by spectral nulls or additive non-speech contributions (e.g. background noise or music). It is demonstrated that the proposed all-pole estimation can significantly improve the performance of speech coders based on sinusoidal model, since the harmonic magnitudes are modeled better by the WMSE all-pole model. 1.

In this paper several techniques are investigated for reduction of complexity and/or improving quality of a line spectrum frequencies (LSF) quantization based on switched prediction (SP) and vector quantization (VQ). For switched prediction, a higher number of prediction matrices is proposed. Quality of the quantized speech is improved by the prediction multi-candidate and delayed decision algorithm. It is shown that quantizers with delayed decision can save up to one bit still having similar or even lower complexity than the baseline quantizers with 2 switched matrices. By efficient implementation of prediction, lower complexity can be achieved through use of prediction matrices with reduced number of non-zero elements. By combining such sparse matrices and multiple prediction candidates, the best quality-complexity compromise quantizers can be obtained as demonstrated by experimental results.

The paper presents a low complexity method for line spectrum frequencies (LSF) coding based on predictive vector quantization (PVQ) with switched-adaptive prediction. Complexity reduction is achieved in two ways. Vector quantization (VQ) is simplified by increasing the number of switched matrices. Secondly, the technique for reduction of the number of nonzero elements in switched matrices is applied to simplify the prediction. The structures of such sparse matrices as well as the optimal values of their nonzero elements are determined to maximize the closed-loop prediction gain of the coder. Designed coders are compared to the baseline coders with full prediction matrices as well as to coders incorporating diagonal matrices. The trade-off between complexity and quality of the proposed coders has been analyzed as a function of various coder parameters, verifying the effectiveness of the proposed approach.

This study deals with the relation between the spectral representation and the perceptual identification of the Spanish fricatives and affricates. Several spectral representations have been analyzed: FFT-derived linear cepstrum, mel cepstrum coefficients, LPC cepstral coefficients and the first four statistical moments. Quadratic discriminant analysis including the leaveone-out method have been carried out on a large database. For this particular classification procedure, both the order of every spectral parametrization and the order of the temporal trajectory of those parameters have been optimized. The results indicate that a low order representation performs satisfactorily and that a three order temporal trajectory is adequate to encode the dynamics of the fricatives. The best classification rates were obtained by the cepstral (79.5%) and linear cepstrum coefficients (75.2%). They attained a correlation coefficient with respect to the perceptual identification of 0.78 and 0.75, respectively.