Noise Alone Reference Research Articles

Predicting speech intelligibility in hearing-impaired listeners using a physiologically inspired auditory model

This study presents a major update and full evaluation of a speech intelligibility (SI) prediction model previously introduced by Scheidiger, Carney, Dau, and Zaar [(2018), Acta Acust. United Ac. 104, 914-917]. The model predicts SI in speech-in-noise conditions via comparison of the noisy speech and the noise-alone reference. The two signals are processed through a physiologically inspired nonlinear model of the auditory periphery, for a range of characteristic frequencies (CFs), followed by a modulation analysis in the range of the fundamental frequency of speech. The decision metric of the model is the mean of a series of short-term, across-CF correlations between population responses to noisy speech and noise alone, with a sensitivity-limitation process imposed. The decision metric is assumed to be inversely related to SI and is converted to a percent-correct score using a single data-based fitting function. The model performance was evaluated in conditions of stationary, fluctuating, and speech-like interferers using sentence-based speech-reception thresholds (SRTs) previously obtained in 5 normal-hearing (NH) and 13 hearing-impaired (HI) listeners. For the NH listener group, the model accurately predicted SRTs across the different acoustic conditions (apart from a slight overestimation of the masking release observed for fluctuating maskers), as well as plausible effects in response to changes in presentation level. For HI listeners, the model was adjusted to account for the individual audiograms using standard assumptions concerning the amount of HI attributed to inner-hair-cell (IHC) and outer-hair-cell (OHC) impairment. HI model results accounted remarkably well for elevated individual SRTs and reduced masking release. Furthermore, plausible predictions of worsened SI were obtained when the relative contribution of IHC impairment to HI was increased. Overall, the present model provides a useful tool to accurately predict speech-in-noise outcomes in NH and HI listeners, and may yield important insights into auditory processes that are crucial for speech understanding.

NAR estimators of spatial covariance matrices for adaptive array detection

The theory of noise-alone-reference (NAR) power estimation is extended to the estimation of spatial covariance matrices. A NAR covariance estimator insensitive to signal presence is derived. The SNR (signal-to-noise ratio) loss incurred by using this estimator is independent of the input SNR and is less than that encountered with the maximum likelihood covariance estimator given that the same number of uncorrelated snapshots is available to both estimators. The analysis assumes first a deterministic signal. The results are extended and generalized to signals with unknown parameters or random signals. For the random signal case, generalized and quasi-NAR covariance estimators are presented.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Noise power estimation and overlapped signals detection

In order to detect a deterministic signal in nonstationary Gaussian noise with power fluctuations and obtain a constant false alarm rate, a classical matched filter receiver normalized by an optimal noise alone reference (NAR) power estimator is used. For the case of overlapped signals, the matched filter characteristic of providing a maximum likelihood estimate of arrival times and signal amplitudes, is lost. The NAR property of the power estimation is at the same time strongly degraded. In this paper a realizable inverse filter of the received signal is derived. By using this filter a high resolution of superposed signals is obtained. Besides, the noise power estimation at the inverse filter output posses the interesting NAR property, even in presence of overlapped signals. Um ein deterministisches Signal in nichtstationärem Gauβschen Rauschen mit Leistungsfluktuationen zu detektieren und dabei eine konstante Falschalarmrate zu erhalten, wird ein klassischer Empfänger mit signalangepaβtem Filter eigensetzt, der durch einen optimalen Schätzer für die Rauschleistung (NAR) normalisiert wird. Für den Fall sich überlappender Signale verliert das signalangepaβte Filter die Eigenschaft, ein maximum Likelihood Schätzer der Ankunftszeiten und Signalamplituden zu sein. Die NAR Eigenschaft des Leistungschätzers für die Rauschleistung wird gleichzeitig sehr abgeschwächt. In diesem Beitrag wird ein realisierbares inverses Filter für das empfangene Signal hergeleitet. Durch den Einsatz dieses Filters wird eine hohe Auflösung der überlappenden Signale erreicht. Die Rauschleistungsschätzung am Ausgang des inversen Filters besitzt auβerdem die interessante NAR Eigenschaft, sogar bei überlappenden Signalen. La détection de signaux déterministes en présence de bruit non-stationnaire, et présentant en particulier des fluctuations de puissance, conduit à introduire des esimateurs de puissance à référence bruit seul (RBS) à l'entrée du filtre adapté afin de réguler la puissance du bruit et d'obtenir une probabilié de fausse-alarme constante. Dans le cas des signaux superpos'es le filtre adapté ne donne plus les estimées du maximum de vraisemblance de l'amplitude et du temps d'arrivée et les propriétés RBS de l'estimateur de puissance est, en méme temps, fortement dégradée. Dans ce travail un filtrage inverse au signal est utilisée pour permettre la discrimination entre signaux adjacents et pour obtenir une estimation de puissance à RBS même en présence des signaux superposés.

Noncoherent NAR power estimators for adaptive detection

Quadratic noise-alone-reference (NAR) power estimators in the form of finite impulse response filters operating on square-law demodulated samples are analyzed. A generalized NAR estimator is obtained by constraining the standard estimator to be nonnegative and unbiased to a symmetrically observed signal. A quasi-NAR estimator is derived to be nonnegative, to be unbiased in the absence of a signal, and to minimize both the variance due to noise alone and the bias due to a symmetrically observed signal. This estimator depends on the signal's normalized envelope but not on the amplitude. The derived estimators are used in determining the structures of two noncoherent adaptive signal detectors in an unknown level additive noise. For independent noise samples, the first detector reduces to a uniform integrator with a cell-averaging constant-false-alarm-rate circuit, and the second reduces to a filter matched to the squared signal envelope but with quasi-NAR automatic gain control, which can regulate faster noise power fluctuations.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Comments on "An axiomatic construction of quadratic NAR power estimators for adaptive detection" [with reply

An unstated axiom in the abovementioned correspondence by M.H. El-Ayadi (see ibid., vol.ASSP-35, p.1347-50, Sept.1987) on quadratic noise estimators requires the kernel matrix to be nonnegative definite. The commenter shows that this assumption is not required for the noise-alone-reference (NAR) estimator; it limits the performance of the generalized NAR estimator in the sense of a minimum-variance unbiased statistic, and it provides the basis of a compromise between variance and bias used to derive the quasi-NAR estimator. It is observed that the nonnegative definite assumption is only one way of introducing a priori information as to the strictly positive nature of power. While this is a reasonable approach, it is not the only one, and it may be possible to realize improved performance in some cases by utilizing other means to introduce the a priori information together with application-specific optimization criteria. The author disagrees with the commenter and presents an argument substantiating his position. He points out that the nonnegative definite constraint is the only approach that makes the class of quadratic estimators considered take on nonnegative values for all possible observation vectors and that removing it does not necessarily result in smaller variance. He contends that the commenter deemphasizes the effect of estimator bias on detection performance. >

Geometrical properties of optimal Volterra filters for signal detection

Linear-quadratic filters are a special example of Volterra filters that are limited to the second order. It is shown that all the results recently published which are valid in the linear-quadratic case can be extended with the appropriate notations to Volterra filters of arbitrary order. Particularly, the optimum Volterra filter giving the maximum of the deflection for detecting a signal in noise is wholly calculated. In addition, several geometrical properties of optimal Volterra filters are investigated by introducing appropriate scalar products. In particular, the concept of space orthogonal to the signal and the noise alone reference (NAR) property are introduced, allowing a decomposition of the optimal filter that exhibits a relation between detection and estimation. Extensions to the infinite case and relations with the likelihood ratio are also investigated. >

Generalized likelihood adaptive detection of signals deformed by unknown linear filtering in noise with slowly fluctuating power

The adaptive detection of signals deformed by unknown linear filtering in noise with slowly fluctuating power is considered. A generalized likelihood ratio test is derived. This test leads to two AGC receivers having the same detection performance. The first receiver is comprised of a classical AGC that cannot distinguish signal from noise and limits the amplitudes of strong signals. The other receiver incorporates a noise-alone reference (NAR) AGC which is insensitive to strong signals and guarantees a better contrast and more reliable amplitude classification. The output contrasts of the two receivers and of a third hypothetical perfect measurement likelihood ratio receiver are compared by means of computer simulations for different signal waveforms and different impulse responses of the linear filter deforming the signal. These simulations show that the performance of the second receiver is quite near to the ultimate performance achieved only theoretically by the third receiver. Moreover, unlike the quasi-NAR AGC receiver with the generalized matched and antimatched filters which was defined in a previous paper, the NAR property of the second receiver is not established at the expense of an increased false alarm probability.

NAR AGC adaptive detection of signals deformed by random linear filtering using generalized matched and antimatched filters

The adaptive detection of signals deformed by random linear filtering in noise with fluctuating power is considered. An automatic gain control (AGC) receiver comprising both generalized matched and antimatched filters is proposed. The antimatched filter is inserted in the AGC channel in order to minimize the average power signal-to-noise ratio, and thus renders the AGC approximately insensitive (transparent) to strong signals, which is the property of a noise alone reference (NAR). On the other hand, the generalized matched filter is used to maximize the ratio of the average received signal power to the average noise power which has been regulated by the AGC. Computer simulation results show cearly the good performance of the proposed AGC receiver.

NAR AGC adaptive detection of nonoverlapping signals in noise with fluctuating power

In order to detect a deterministic signal in nonstationary noise with slow power fluctuations and obtain a constant false alarm rate, an automatic gain control (AGC) system is generally used at the input of the matched filter. But this AGC system does not differentiate between signal and noise and, in general, limits the amplitudes of strong signals. In this paper we present another AGC system possessing an interesting property called the noise alone reference (NAR) property. It means that the NAR AGC system acts only on the noise even in the presence of signal. The performance of the NAR AGC system is compared with that of the classical one in the presence of localized signals (as in communication problems) or unlocalized signals (as in radar and sonar problems). We show that the NAR AGC system provides better contrast and more reliable amplitude classification at the output of the receiver. Finally, computer simulation results are presented to show the effects of NAR AGC systems in adaptive detection.