Binaural Signal Integration Improves Vertical Sound Source Localization

Early studies have shown that the localization of a sound source in the vertical plane can be accomplished with only a single ear and thus assumed to be based on monaural spectral cues. Such cues consists of notches and peaks in the perceived spectrum which vary systematically with the elevation of sound sources. This poses several problems to the auditory system like extracting relevant and direction-dependent cues among others. Interestingly, at the stage of elevation estimate binaural information from both ears is already available and it seems reasonable of the auditory system to take advantage of this information. Especially, since such a binaural integration can improve the localization performance dramatically as we demonstrate with a computational model of binaural signal integration for sound source localization in the vertical plane. In line with previous findings of vertical localization, modeling results show that the auditory system can perform monaural as well as binaural sound source localization given a single, learned map of binaural signals. Binaural localization is by far more accurate than monaural localization, however, when prior information about the perceived sound is integrated localization performance is restored. Thus, we propose that elevation estimation of sound sources is facilitated by an early binaural signal integration and can incorporate sound type specific prior information for higher accuracy.

Deep neural network models of sound localization reveal how perception is adapted to real-world environments

Mammals localize sounds using information from their two ears. Localization in real-world conditions is challenging, as echoes provide erroneous information, and noises mask parts of target sounds. To better understand real-world localization we equipped a deep neural network with human ears and trained it to localize sounds in a virtual environment. The resulting model localized accurately in realistic conditions with noise and reverberation, outperforming alternative systems that lacked human ears. In simulated experiments, the network exhibited many features of human spatial hearing: sensitivity to monaural spectral cues and interaural time and level differences, integration across frequency, and biases for sound onsets. But when trained in unnatural environments without either reverberation, noise, or natural sounds, these performance characteristics deviated from those of humans. The results show how biological hearing is adapted to the challenges of real-world environments and illustrate how artificial neural networks can extend traditional ideal observer models to real-world domains.

Modeling perceived externalization of a static, lateral sound image

Perceived externalization is a relevant feature to create an immersive acoustic environment with headphone reproduction. In the present study, listener-specific acoustic transfer characteristics for an azimuth angle of 90° were modified to investigate the role of monaural spectral cues, interaural level differences (ILDs), and temporal fluctuations of ILDs on perceived externalization in anechoic and reverberant environments. Listeners’ ratings suggested that each acoustic cue was important for perceived externalization. If only one correct acoustic cue remained in the ear signals, the sound image could not be perceived as fully externalized. Reverberation did reduce but not eliminate the influences of monaural spectral and ILD cues on perceived externalization. Additionally, the spectral details of the ipsilateral ear signal were more important for perceived externalization than those in the contralateral ear signal. A computational model was proposed to quantify those relationships and predict externalization ratings by comparing the acoustic cues extracted from the target (modified) and template (non-processed) binaural signals after several auditory processing steps. The accuracy of predicted externalization ratings was higher than 90% under all experimental conditions.

Sound Localization Under Perturbed Binaural Hearing

This paper reports on the acute effects of a monaural plug on directional hearing in the horizontal (azimuth) and vertical (elevation) planes of human listeners. Sound localization behavior was tested with rapid head-orienting responses toward brief high-pass filtered (>3 kHz; HP) and broadband (0.5–20 kHz; BB) noises, with sound levels between 30 and 60 dB, A-weighted (dBA). To deny listeners any consistent azimuth-related head-shadow cues, stimuli were randomly interleaved. A plug immediately degraded azimuth performance, as evidenced by a sound level–dependent shift (“bias”) of responses contralateral to the plug, and a level-dependent change in the slope of the stimulus–response relation (“gain”). Although the azimuth bias and gain were highly correlated, they could not be predicted from the plug's acoustic attenuation. Interestingly, listeners performed best for low-intensity stimuli at their normal-hearing side. These data demonstrate that listeners rely on monaural spectral cues for sound-source azimuth localization as soon as the binaural difference cues break down. Also the elevation response components were affected by the plug: elevation gain depended on both stimulus azimuth and on sound level and, as for azimuth, localization was best for low-intensity stimuli at the hearing side. Our results show that the neural computation of elevation incorporates a binaural weighting process that relies on the perceived, rather than the actual, sound-source azimuth. It is our conjecture that sound localization ensues from a weighting of all acoustic cues for both azimuth and elevation, in which the weights may be partially determined, and rapidly updated, by the reliability of the particular cue.

Directionality Derived From Differential Sensitivity to Monaural and Binaural Cues in the Cat's Medial Geniculate Body

Azimuth tuning of high-frequency neurons in the primary auditory cortex (AI) is known to depend on binaural disparity and monaural spectral (pinna) cues present in broadband noise bursts. Single-unit response patterns differ according to binaural interactions, strength of monaural excitatory input from each ear, and azimuth sensitivity to monaural stimulation. The latter characteristic has been used as a gauge of neural sensitivity to monaural spectral directional cues. Azimuth sensitivity may depend predominantly on binaural disparity cues, exclusively on monaural spectral cues, or on both. The primary goal of this study was to determine whether each cortical response pattern corresponds to a similar pattern in the medial geniculate body (MGB) or whether some patterns are unique to the cortex. Single-unit responses were recorded from the ventral nucleus (Vn) and lateral part of the posterior group of thalamic nuclei (Po), tonotopic subdivisions of the MGB. Responses to free-field presentation of noise bursts that varied in azimuth and sound pressure level were obtained using methods identical to those used previously in field AI. Many units were azimuth sensitive, i.e., they responded well at some azimuths, and poorly, if at all, at others. These were studied further by obtaining responses to monaural noise stimulation, approximated by reversible plugging of one ear. Monaural directional (MD) cells were sensitive to the azimuth of monaural noise stimulation, whereas binaural directional (BD) cells were either insensitive to its azimuth or monaurally unresponsive. Thus BD and MD cells show differential sensitivity to monaural spectral cues. Monaural azimuth sensitivity could not be used to interpret the spectral sensitivity of predominantly binaural cells that exhibited strong binaural facilitation because they were either unresponsive or poorly responsive to monaural stimulation. The available evidence suggests that some such cells are sensitive to spectral cues. The results do not indicate the presence of any response types in AI that are not present in the MGB. Vn and Po contain similar classes of MD and BD cells. Because Po neurons project to the anterior auditory field, neurons in this cortical area also are likely to exhibit differential sensitivity to binaural disparity and monaural spectral cues. Comparison of these MGB data with a published report of cochlear nucleus (CN) single-unit azimuth tuning shows that MGB sensitivity to spectral cues is considerably stronger than CN sensitivity.

Altered Spectral Localization Cues Disrupt the Development of the Auditory Space Map in the Superior Colliculus of the Ferret

Schnupp, Jan W. H., Andrew J. King, and Simon Carlile. Altered spectral localization cues disrupt the development of the auditory space map in the superior colliculus of the ferret. J. Neurophysiol. 79: 1053–1069, 1998. Spectral localization cues provided by the outer ear are utilized in the construction of the auditory space map in the superior colliculus (SC). The role of the outer ear in the development of this map was examined by recording from the SC of anesthetized, adult ferrets in which the pinna and concha had been removed in infancy. The acoustical consequences of this procedure were assessed by recording outer ear impulse responses via a probe-tube microphone implanted in the wall of the ear canal. Both monaural and binaural spectral cues normally show a number of asymmetric features within the horizontal plane, which allow azimuthal locations on either side of the interaural axis to be discriminated. These features were eliminated or altered by chronic pinnectomy. The responses of auditory units in the SC to noise bursts presented in the free field were examined at sound levels of ∼10 and 25 dB above unit threshold. After bilateral pinnectomy, the representation of auditory space was severely degraded at both sound levels. In contrast to normal ferrets, many units had bilobed azimuthal response profiles, indicating that they were unable to resolve sound locations on either side of the interaural axis. There was also much less order in the distribution of best azimuths or elevations of those units that were tuned to a single direction. Some units were tuned to locations that extended much further into the hemifield ipsilateral to the recording side than the normal range of best azimuths. Unilateral removal of the outer ear, which disrupts the monaural spectral cues for one side only, had a much smaller effect on the development of the auditory representation. At supra- and near-threshold sound levels, the representation of sound azimuth in the SC on both sides of the brain was less scattered than that found after bilateral pinna removal. Nevertheless, units with bilobed responses, broader tuning, and inappropriate best azimuths were observed in both the left and right SC of ferrets in which the left pinna and concha had been removed in infancy. These data illustrate that the localization cues provided by the outer ear play a critical role in the development of the auditory space map in the SC. In contrast to other manipulations of either auditory or visual inputs, the map does not appear to adapt to the changes in spectral cues brought about by pinna removal, suggesting that residual binaural cues are, by themselves, insufficient for its normal maturation.