Phase Sensitivity Of A Computer Model Of The Auditory Periphery - Research Paper Example

 Phase Sensitivity Of A Computer Model Of The Auditory Periphery Summarize and Paraphrase 1. Pitch Identification 1. a. Autocorrelation Timing information from the auditory nerve activity is useful in pitch identification. Models of pitch identification based on the extraction of this timing information exist. Licklider 1951, suggested such a model employing autocorrelation analysis. Licklider 1951, was the first to propose the use of autocorrelation analysis in pitch identification based on the timing information of auditory nerve activity. (1). Humans possess the ability to perceive the pitch present in a complex tone, though the tone does not contain any spectral component that has a correspondence with the relevant pitch. The prevalent theories at that time posted that the phenomenon was fully explained by the distortion products that resulted from the cochlear responses. Licklider 1951 rejected this explanation and based his theory on the definite possibility of using the waveform envelope of unresolved harmonic components for the extraction of pitch information by the use of autocorrelation analysis. The performance of such an autocorrelation analysis could be done using a delay line mechanism on the auditory nervous system at a low level. The model proposed by Licklider 1951 posits that it is possible to extract pitch from eighth nerve firing patterns by employing a running autocorrelation function, which is conducted on individual fibre activity. (1). The theory of Licklider 1951, is thus essentially based on the concept of the harmonic components not being fully resolved, which enabled the period of the pitch to be represented by the periodicity seen in the firing of the individual auditory-nerve fibres. Later studies however made it clear that the most important pitch perceptions were founded on the complex stimuli, wherein the individual harmonic components were spaced widely apart most surely resolved by the auditory fibres, in a manner whereby the periodicity of individual will not be a proper guide to the pitch of the stimulus as believed by Licklider 1951. (1). The presence of this clarification resulted in theories that concentrated on spectral cues to pitch that were founded mostly on the resolved harmonics of the tone complex. Though these theories were suitable in providing an explanation for the results of psychophysical experiments, they were found deficient in the case of non-spectral pitch phenomena, like those stemming from periodically interrupted or amplitude-modulated noise. Subsequent developments in the field have shown that unresolved harmonics are a contributory factor to pitch identification along with resolved harmonics and hence the combination of both the sources is necessary for the generation of the pitch perception. (1). Furthermore there is the issue of the relative merits relating to the spectral and temporal analysis of such stimuli, though there is general acceptance of mechanical to neural trans-reduction process that happens at the cochlea impacts on the limited frequency analysis of the signal. The “Place Theories” perceive this resolution as the starting point for the pattern recognition process in which pitch is linked to a pattern of activity that is spread across several sites on the basilar membrane. In contrast to this the temporal theories, like Licklider’s autocorrelation analysis, uses analysis focused on the time intervals among the spikes seen in the individual AN fibres. In this study the model used consists of autocorrelation on the activity of sets of simulated nerve fibres and then aggregates the autocorrelation functions (ACF) obtained to provide a summary autocorrelation function. The study aims at showing that this summary ACF has the pertinent information that assists the aim of simulating human listener performance over a broad range of psychophysical studies. (1). 1. b. Summary of the Study The model used in the study consists of the following set of activities. Outer-ear frequency bandpass function; low-and high-frequency attenuation of the middle-ear; mechanical filtration of the basilar membrane; mechanical to neural transduction at the site of the hair cell; refraction inhibition of firing as demonstrated by the auditory-nerve fibres; approximation of the distribution in the intervals among the spikes that are found to start from fibres along the same channel; summing of these interval estimates that span the channels; and extracting the pitch through the scrutiny of the summary ACF. (1). Sampling of the signal to be used was done and the model was updates 20,000 times/sec. Starting from stage four of the activities the process was described with respect to spikes and the interval between these spikes, but the computation was done exclusively with regard to the probability of the occurrence of spikes. The measurement of the time intervals among the spikes was in reference to the time interval that occurred between every spike and every other spike that was seen in the same channel. The study employed the convention that a signal rms of 1 be taken as 0 dBl and on the basis of the scale employed in the study the auditory-nerve fibre was described as having a threshold of 15 dBl. (1) The model of auditory periphery as given by Licklider 1951, suggests that the auditory-nerve fibre spike intervals should be aggregated such that it has mathematical similarity to autocorrelation functions. The study uses such a model, with a novel addition of the principle of aggregating information spread over frequency-selective channels. The addition of this novel feature makes the Licklider, 1951 model more satisfactory in terms of being a model of human listener’s pitch perception. (1). The development of a model using the time interval approach is hampered by the lack of clarity in the understanding of the physiology of the timing information and identification of the nucleus in the nervous system responsible for it. In spite of these hurdles it is possible to assume that the nervous system extracts pitch based on timing information and then pools this information across the filtered channels. (1). Though the model is has computational intensive properties, the manner in which handling of the auditory-nerve spikes occurs is very simple, wherein there is just the requirement for aggregation across channels the time intervals among spikes within the channels. This simple principle is responsible for the properties of the model and compares well against the existent pitch theories. (1). Time-Domain Modelling A functional model of the cochlear capable of simulating the phase-locked activity similar to which complex sounds generate in the auditory nerve had been developed by Patterson et al, termed the auditory image model (AIM). The purpose of the study was to evaluate the role played by fine-grain timing information on auditory perception in general and in the perception of speech in particular. (2). In the AIM there are two routes consisting of the functional route and the physiological route in middle ear filtering, with three similar functions of different implementation, producing different simulations. The functional route of middle ear filtering is made up of the stages of spectral analysis that is implemented through gammatone filtering to produce the simulation of basilar membrane motion; the neural encoding stage that is implemented through compression and two-dimensional adaptive thresholding to give the simulation of neural activity pattern and the time-interval stabilization stage that is implemented through strobed temporal integration to produce the simulation of the auditory image. The physiological route of middle ear filtering is made up of the spectral analysis stage, where implementation is through the transmission line filtering, spectral sharpening and compression, to produce the simulation of basilar membrane motion; neural encoding stage, wherein implementation is through inner hair cell simulation to produce the simulation of neural activity and the time interval stabilization stage, where autocorrelation is the implementation mode, to produce the correlogram. Such an auditory image model makes it possible to generate sequences of simulated images at regular time intervals in the form of an animated cartoon in which the dynamic behaviour of the auditory images generated by common sounds can be shown. (2). Users of the AIM may use the physiological model or the functional model to simulate outputs required for their purposes or switch between the two models. Such switching is made possible by the switches incorporated that permit the use to switch from the functional model to the physiological model at the output stage of each model. (2). In the spectral analysis stage, spectral analysis is done with the help of a bank of auditory filters, which perform the function of converting a digitized wave into an array of filtered waves. This set of waves represents basilar membrane motion in the AIM. The software program of the module distributes the filters in a linear pattern along the frequency scale that is measured in equivalent rectangular bandwidths (ERBs). Options are available that enable the user to choose the number of channels used in the filterbank and also the minimum and maximum filter centre frequencies. The module has functional auditory filter as well as the physiological auditory filter for producing the BMM. The functional auditory filter is a gammatone filter, while the physiological auditory filter is a nonlinear transmission filter. (2) The neuronal encoding stage is the second stage of AIM, wherein mechanical or neuronal transduction processes of the inner hair cells are simulated. In this simulation BMM is converted into NAP. There are two alternative simulations are available for producing NAP, which are the bank of two-dimensional thresholding units and the bank of inner hair cell simulators. (2) Static perceptions rather than oscillating perceptions are produced by periodic sounds, which give rise to the suggestions that temporal integration is applied to the NAP in the production of the human initial perception of sound or the human auditory image. A simp e leaky integration process can be used to represent auditory temporal integration, but there is the problem that this will remove the phase-locked fine structure seen with NAP, which would be in conflict with perceptual data that indicates the important role played by fine structure in the determination of sound quality and source identification. To overcome this AIM makes use of two modules that retain most of the time-interval information in NAP at the time of temporal integration and this produces a better representation of human auditory images. In the functional mode of AIM, strobed temporal integration is used to achieve this, while in the physiological mode a bank of auto-correlators are used. (2). Common Factors in the Identification of Brief Everyday Sounds There is limited clarity in the manner in which humans perceive and identify everyday sounds. This may sound surprising in the face of the regular presence and the importance of their functional role. Ballas 1993 conducted a set of experiments using a set of sounds to collect acoustic, perceptual and cognitive data for evaluating their role in the identification of brief everyday sounds. (3). The first experiment was to done to throw more light on the causal uncertainty and the identification response in times for the set of sounds. Based on the results of the experiments Ballas 1993 confirms that there is firm monotonic relationship between identification time and Hcu. Furthermore LMIT is better estimated with the assistance of Hcu than through the use of acoustic measures computed for the sounds, but the combination of acoustic measures with Hcu provide a better picture in accounting for nearly half of the variances seen in identification time. (3). In the second experiment Ballas 1993, sought to explore the occurrence frequency of the sounds that were used in the earlier experiment. Most of the sounds used were heard from 10.00 am to 7.00 pm, however it was the period 8.00 pm and midnight that a greater proportion of the sounds were reported. These sounds were heard with almost the same frequency during weekdays and Saturday was when the least number of sounds of sound were reported. Nearly half of the sounds were heard at home as a result of internal or external sounds. The sounds were also heard at the workplace, grocery store or at airports or while exercising in a gym, movie theatre or restaurant and the like. As a result of the findings of the study Ballas 1993 suggests that performance on a group of varied everyday sounds is best related to a group of measures for different environments rather than using a single measure for these sounds. (4). The third experiment conducted by Ballas 1993, was on the role played by perceptual and cognitive judgements on the perceptual and cognitive processed involved in sound identification. Based on the findings Ballas 1993, sums up that cognitive and perceptual judgements plays a role in sound identification and that knowledge representation of everyday sounds may be founded on perceptual factors and yet reflect the phenomena of the events. (4). Having established that acoustic, perceptual and cognitive factors are important to the sound identification process, Ballas 1993 then proceeded to evaluate sound labelling and causal uncertainty in sound identification in experiment 5 and experiment 6 Ballas 1993, found that causal uncertainty drawn from labelling correlated highly with identification time and accuracy and had a weak relationship with ecological frequency. (4). Works Cited 1. Meddis, Ray & Hewitt, J. Michael. “Virtual pitch and phase sensitivity of a computer model of the auditory periphery. I: Pitch identification”. Journal of the Acoustic Society of America 89.6 (1991): 2866-2882. 2. Patterson, D. Roy & Allerhand, H. Mike. “Time-domain modelling of peripheral auditory processing: A modular architecture and a software platform”. Journal of the Acoustic Society of America 98.4 (1995): 1890-1894. 3. Ballas, A. James. “Common Factors in the Identification of an assortment of Brief Everyday Sounds”. Journal of Experimental Psychology: Human Perception and Performance. 19.2 (1993): 250-267. Read More