Estimates of Human Cochlear Tuning at Low LevelsUsing Forward and Simultaneous MaskingANDREWJ. OXENHAM,1AND CHRISTOPHERA. SHERA2,31Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA2Eaton–Peabody Laboratory of Auditory Physiology, Massachusetts Eye and Ear Infirmary, Boston, MA 02114, USA3Department of Otology and Laryngology, Harvard Medical School, Boston, MA 02115, USAReceived: 14 November 2002; Accepted: 9 May 2003; Online publication: 10 July 2003ABSTRACTAuditory filter shapes were derived from psycho-physical measurements in eight normal-hearing lis-teners using a variant of the notched-noise methodfor brief signals in forward and simultaneous mask-ing. Signal frequencies of 1, 2, 4, 6, and 8 kHz weretested. The signal level was fixed at 10 dB above ab-solute threshold in the forward-masking conditionsand fixed at either 10 or 35 dB above absolutethreshold in the simultaneous-masking conditions.The results show that filter equivalent rectangularbandwidths (ERBs) are substantially narrower in for-ward masking than has been found in previous stud-ies using simultaneous masking. Furthermore, incontrast to earlier studies, the sharpness of tuningdoubles over the range of frequencies tested, givingQERBvalues of about 10 and 20 at signal frequenciesof 1 and 8 kHz, respectively. It is argued that the newestimates of auditory filter bandwidth provide a moreaccurate estimate of human cochlear tuning at lowlevels than earlier estimates using simultaneousmasking at higher levels, and that they are thereforemore suitable for comparison to cochlear tuning datafrom other species. The data may also prove helpfulin defining the parameters for nonlinear models ofhuman cochlear processing.Keywords: cochlear tuning, psychoacoustics,auditory filters, maskingINTRODUCTIONThe ability to perceptually separate simultaneoussounds of different frequencies is a fundamentalproperty of the auditory system. The filtering ofsounds according to frequency and the resultingtonotopic organization found throughout the audi-tory pathways to the auditory cortex (Read et al.2002) have their basis in the electromechanicalproperties of the cochlea. Indeed, it is widely believedthat frequency selectivity, measured behaviorally, canbe regarded as a direct reflection of the filtering thattakes place in the cochlea.Estimates of frequency selectivity in humans haveevolved over many decades as numerous confound-ing factors and potential artifacts were addressed(e.g., Wegel and Lane 1924; Fletcher 1940; Zwicker etal. 1957; Bos and de Boer 1966; Houtgast 1973; Pat-terson and Nimmo–Smith 1980; Moore et al. 1984;Glasberg and Moore 2000). The psychophysical tun-ing curve (PTC) (Houtgast 1973; Moore 1978) pro-vides a paradigm that is, in principle, most similar tophysiological measures of cochlear tuning, such asthe neural tuning curve. However, issues such as off-frequency listening and the detection of beats and/ordistortion products can make the results from PTCexperiments difficult to interpret (Johnson–Daviesand Patterson 1979; O’Loughlin and Moore 1981;Patterson and Moore 1986). Over the last two dec-ades, the notched-noise technique has become thefavored method of behaviorally estimating frequencyselectivity (Patterson 1976; Patterson and Nimmo–Smith 1980; Moore 1987; Rosen et al. 1998; Glasbergand Moore 2000). This technique involves measuringCorrespondence to: Andrew J. Oxenham, Ph.D. MITÆbldg. 36-763ÆCambridge, MA 02139ÆTelephone: (617) 253-5995; fax: (617) 258-7003; email: [email protected] 4: 541–554 (2003)DOI: 10.1007/s10162-002-3058-yJAROJournal of the Association for Research in Otolaryngologythe masked threshold of a sinusoidal signal in thepresence of a noise with a spectral notch as a functionof the width and position of the notch relative to thesignal. The filter functions derived from such tech-niques are often referred to as auditory filters and arebelieved to reflect the filtering properties of thecochlea (Moore 1995). One of the most compre-hensive studies of frequency selectivity using notched-noise maskers provides a function that can be used tocalculate the estimated equivalent rectangular band-width (ERB) of the auditory filter at any given fre-quency (Glasberg and Moore 1990). The equation forthe function isERB ¼ 24:7ð4:37f =1000 þ 1Þð1Þwhere ERB is the equivalent rectangular bandwidthand f is the filter center frequency, both in Hz. Thisfunction has been used as an estimate of humancochlear tuning in a wide range of studies and ap-plications (e.g., Beauvois and Meddis 1996; Dau et al.1996, 1997; Moore et al. 1997; Breebaart et al. 2001).In all these studies, the auditory filters are assumed tobe linear in any given condition, although it is ex-plicitly acknowledged that they change their shapewith level.While ignoring cochlear nonlinearities for the sakeof simplicity has certain advantages, some importanteffects are dependent on these nonlinearities. Oneconsequence of cochlear nonlinearity is known assuppression, whereby the neural response to onetone can be reduced by the introduction of a second,suppressor, tone (Sachs and Kiang 1968). It has beenknown since this effect was first investigated behavi-orally that estimates of cochlear tuning can be alteredby effects ascribed to suppression. In general, esti-mates using simultaneous masking, where the maskeris thought to suppress the signal to some degree,produce wider estimates of filter bandwidth than doestimates using nonsimultaneous masking, where themasker does not peripherally interact with, or sup-press, the signal (Houtgast 1973, 1974; Moore 1978;Vogten 1978; Moore et al. 1987). Consistent withthese findings, Heinz et al. (2002) showed in a recentmodeling study that when their nonlinear model wasused to predict thresholds in a simultaneous maskingnotched-noise experiment, the resulting estimatedauditory filter bandwidths were wider than thebandwidths of the filters actually used in the model.In other words, the nonlinearities in the model incombination with the simultaneous notched-noisemethod resulted in biased estimates of filter band-width. As their model incorporated stronger nonlin-earity at high frequencies than at low frequencies (inline with relevant physiological data), the bias effectsincreased with increasing center frequency.The prediction that suppression may affect tuningestimates more at high characteristic frequencies(CFs) than at low CFs is