Abstract
Hearing impairment is often associated with damage to the hair cells in the cochlea. Sometimes there may be complete loss of function of inner hair cells (IHCs) over a certain region of the cochlea; this is called a “dead region”. The region can be defined in terms of the range of characteristic frequencies (CFs) of the IHCs and/or neurons immediately adjacent to the dead region. This paper reviews the following topics: the effect of dead regions on the audiogram; methods for the detection and delineation of dead regions based on psychophysical tuning curves (PTCs) and on the measurement of thresholds for pure tones in “threshold equalizing noise” (TEN); effects of dead regions on speech perception; effects of dead regions on the perception of tones; implications of dead regions for fitting hearing aids. The main conclusions are: (1) Dead regions may be relatively common in people with moderate-to-severe sensorineural hearing loss; (2) Dead regions cannot be reliably diagnosed from the audiogram; (3) PTCs provide a useful way of detecting dead regions and defining their boundaries. However, the determination of PTCs is probably too time-consuming to be used for routine diagnosis of dead regions in clinical practice; (4) The measurement of detection thresholds for pure tones in TEN provides a simple method for clinical diagnosis of dead regions; (5) Pure tones with frequencies falling in a dead region do not evoke clear pitch sensations (pitch matching is highly variable) and the perceived pitch is sometimes, but not always, different from “normal”. However, ratings of pitch clarity cannot be used as a reliable indicator of a dead region; (6) Amplification of frequencies well inside a high-frequency dead region usually does not improve speech intelligibility, and may sometimes impair it. However, there may be some benefit in amplifying frequencies up to 50 to 100% above the estimated low-frequency edge of a high-frequency dead region; (7) The optimal form of amplification for people with low-frequency dead regions remains somewhat unclear. There may be some benefit from avoiding the amplification of frequencies well inside a dead region; (8) Patients with extensive dead regions are likely to get less benefit from hearing aids than patients without dead regions; (9) For patients with diagnosed dead regions at high frequencies, consideration should be given to use of a hearing aid incorporating frequency transposition and/or compression.
Introduction and Definitions
It has been known for many years that cochlear hearing loss is sometimes associated with complete destruction of the inner hair cells (IHCs) within the cochlea (Engström, 1983). For a review, see Borg et al. (1995). Sometimes the IHCs may still be present, but may be sufficiently abnormal that they no longer function. The IHCs are the transducers of the cochlea, responsible for converting the vibration patterns on the basilar membrane into action potentials in the auditory nerve. When the IHCs are non-functioning over a certain region of the cochlea, no transduction will occur in that region. For this reason, I will refer to such a region as a dead region.
If one or more dead regions are present in a specific patient, this has strong implications for the way that a hearing aid should be chosen and fitted for that patient. It can also have strong implications for the extent of benefit that might be gained from a hearing aid. These points will be discussed in more detail later in this paper. Firstly, I consider the ways that dead regions can be characterized and measured.
One way of characterizing a dead region is in terms of the place in the cochlea that is dead. For example, one might refer to a basal dead region or an apical dead region. An alternative definition is in terms of the range of characteristic frequencies (CFs) of the IHCs or neurons that would normally be associated with that region. Say, for example, that the IHCs are non-functioning over a region of the basilar membrane where the IHCs and neurons normally have CFs in the range 4000 to 10000 Hz. One might then describe this as a dead region extending from 4000 to 10000 Hz. This definition was adopted in an earlier paper (Moore et al., 2000b). However, one problem associated with this definition lies in the use of the word “normally”; in an ear with cochlear hearing loss, damage to the outer hair cells (OHCs) can lead to shifts in CF relative to the “normal” values (Liberman and Dodds, 1984). This happens because the CF of the basilar membrane shifts when the function of the OHCs is impaired (Sellick et al., 1982; Ruggero et al., 1997). Since IHC damage is often associated with OHC damage, the tuning of the basilar membrane, IHCs and neurons may be abnormal in an ear with a dead region, even over regions which are not dead.
In the present paper, I adopt a somewhat different definition. A dead region is defined in terms of the CFs of the IHCs and/or neurons immediately adjacent to the dead region. This definition is appropriate even if the CFs of the IHCs and neurons are shifted from “normal” values. The definition also allows for an easy interpretation of the psychoacoustic results that will be presented later in this paper.
It should be noted that the nerve fibers innervating a dead region may degenerate after a time (Friedman, 1997). This is of importance for people who are candidates for a cochlear implant, but it probably has few functional consequences for users of a hearing aid; if the IHCs are nonfunctional in a certain region, the primary auditory neurons innervating that region will play little or no role in perception.
The Role of Off-Frequency Listening
Basilar-membrane vibration in a dead region is not detected via the neurons directly innervating that region. Say, for example, that the IHCs at the basal end of the cochlea are non-functioning. Neurons innervating the basal end, which would normally have high CFs, will not respond. However, if a high-frequency sinusoid is presented, it may be detected if it produces sufficient basilar-membrane vibration at a more apical region; this corresponds to downward spread of excitation. In other words, a high-frequency sound may be detected via neurons that are tuned to lower frequencies. Similarly, if there are no functioning IHCs at an apical region of the cochlea, a low-frequency sound may be detected via neurons that are tuned to higher frequencies, i.e. via upward spread of excitation (Thornton and Abbas, 1980; Florentine and Houtsma, 1983). Because of this possibility, the “true” hearing loss at a given frequency may be greater than suggested by the audiometric threshold at that frequency.
Detection of a tone of a particular frequency via IHCs and neurons with CFs different from that of the tone, is often called “off-frequency listening” (Johnson-Davies and Patterson, 1979; Patterson and Nimmo-Smith, 1980; O'Loughlin and Moore, 1981; Patterson and Moore, 1986), although it would perhaps be more accurately called “off-place listening”. Off-frequency listening can occur even in normally hearing people, especially when a signal has to be detected in the presence of another sound. However, it occurs in a much more extreme form in people with dead regions. The concept of off-frequency listening plays a key role in interpreting psychoacoustical results obtained from people with dead regions.
The Effect of Dead Regions on the Audiogram
It has been recognized for many years that, when a dead region is present, the audiogram will give a misleading impression of the amount of hearing loss, for a tone whose frequency falls in the dead region (Gravendeel and Plomp, 1960; Halpin et al., 1994). Effectively, the “true” hearing loss in a dead region is infinite, but the audiogram may sometimes indicate only a moderate hearing loss. I will consider this point in more detail later on. Firstly, though, I will consider the extent of “true” hearing loss that can occur without a dead region being present.
Effects of Hair-Cell Damage (But Not Death) on the Audiogram
Damage to the hair cells (but without complete loss of function of the IHCs) can give rise to elevated absolute thresholds in two main ways. Firstly, dysfunction of the OHCs impairs the active mechanism in the cochlea, resulting in reduced basilar membrane vibration for a given low sound level (Ruggero, 1992; Yates, 1995; Moore, 1998). Hence, the sound level must be larger than normal to give a just-detectable amount of vibration. Secondly, dysfunction of the inner hair cells (IHCs) can result in reduced efficiency of transduction, so the amount of basilar membrane vibration needed to reach threshold is larger than normal. The dysfunction of the IHCs might occur in several ways: patchy death of IHCs (complete loss of some IHCs), other IHCs functioning near-normally; damage of the majority of IHCs without death of any IHCs; patchy death of IHCs with damage to others; metabolic disturbance; structural problems, such as shrinkage of the tectorial membrane; or a combination of these.
In principle, it is possible to partition the overall hearing loss at a given frequency into a component due to OHC damage and a component due to IHC (and neural) damage (Moore and Glasberg, 1997):
where all quantities are in decibels. For example, if the total hearing loss at a given frequency is 60 dB, 40 dB of that loss might be due to OHC damage and 20 dB to IHC damage. It is not possible to determine the balance between the two components from measures of absolute threshold alone. However, the amount of hearing loss due to OHC damage, HLOHC, can be estimated in other ways, for example from the rate of growth of loudness with increasing sound level (Moore and Glasberg, 1997), from estimates of the sharpness of the auditory filter (Moore et al., 1999b) or from comparison of growth-of-masking functions in forward masking for a masker centered at the signal frequency and a masker centered below the signal frequency (Oxenham and Plack, 1997; Moore et al., 1999b). Note that the proportion of a hearing loss that is attributed to OHC or IHC damage is not the same as the relative proportion of OHC and IHC physiological damage. In the above example, 40 dB of the hearing loss was attributed to OHC damage and 20 dB to IHC damage, but this does not imply that damage to the OHCs was twice as great as damage to IHCs.
It is, of course, possible for the functioning of the hair cells to be reduced without the hair cells themselves being physically damaged. For example, if the metabolism of the cochlea was affected by malfunctioning of the stria vascularis, this could adversely influence the functioning of both OHCs and IHCs. The effect of this for OHCs would be very similar to the effect of damage to the OHCs, namely that the active mechanism would be impaired in its operation, so a larger than normal input sound level would be required to give a just-detectable amount of vibration on the basilar membrane. However, for the IHCs the effect of metabolic disturbance might be different from the effect of damage to the IHCs themselves.
Physiological data obtained from animals suggest that the maximum “gain” provided by the active mechanism is about 50 dB at low frequencies and 65 dB at high frequencies (Yates, 1990; 1995; Ruggero et al., 1997). This represents the difference in input sound level needed to produce a fixed low amount of vibration on the basilar membrane in a healthy ear and an ear in which the active mechanism is not functioning (for example following the death of the animal). Psychophysical data obtained from humans suggest gains of similar magnitude, and also suggest that the gain of the active mechanism is greater at high frequencies than at low frequencies (Oxenham and Plack, 1997; Plack and Oxenham, 1998; Bacon et al., 1999; Hicks and Bacon, 1999; Moore et al., 1999b; Plack and Oxenham, 2000). It seems likely that hearing loss due purely to OHC dysfunction cannot be greater than 50 dB at low frequencies and 65 dB at high frequencies. Therefore, any (cochlear) hearing loss greater than this must arise at least partly from IHC dysfunction, either because of damage to the IHCs, or because of other factors that affect IHC function, such as metabolic disturbance.
The question now arises: how great does the damage to the IHCs have to be before they are considered effectively dead? Put another way, is there a limit to the value of HLIHC, above which there is effectively a dead region? These questions are difficult to answer. There are probably cases where the IHCs are severely damaged, and hence require a larger input than normal to function, but which nevertheless are capable of initiating action potentials in the auditory nerve if the input level is sufficiently high. A value of HLIHC of, say, 20 dB, implies that the basilar membrane vibration has to be 20 dB greater than normal to evoke a threshold response, which corresponds to an amplitude or velocity of vibration ten times as great as normal. A value of HLIHC of 40 dB implies that the amplitude of basilar membrane vibration has to be 100 times as great as normal, which is a rather extreme value.
We can get some idea of when the IHCs become effectively dead from physiological data. Dysfunction of the OHCs (and hence the active mechanism) mainly affects basilar membrane responses to tones with frequencies close to CF (Sellick et al., 1982; Robles et al., 1986; Ruggero et al., 1997). The response to tones with frequencies well below CF is almost unaffected by the integrity of the OHCs (although sometimes sensitivity to such tones can actually be increased by about 10 dB when the OHCs are damaged). Therefore, the effect of IHC dysfunction in a hearing-impaired animal can be quantified by measuring the elevation in threshold (relative to that for a normal animal) for a tone with frequency well below CF. Measurements of this type were performed by Liberman and Dodds (1984) in animals with cochlea damage produced by intense noise or by the ototoxic drug kanamycin. They found that threshold elevations associated with IHC damage were mostly less than 20 dB, and rarely exceeded 30 dB. Neurons showing threshold elevations greater than 30 dB were presumably not found because, beyond that point, the IHCs were effectively dead, and incapable of initiating action potentials.
Another way of assessing the effective upper limit of HLIHC is to examine the largest threshold shift that can be measured in response to a CF tone. Liberman and his co-workers have described an extensive series of studies of neural responses in animals with cochlear damage produced by noise, ototoxic drugs, or a combination of the two (Liberman and Kiang, 1978; Liberman and Dodds, 1984; Liberman et al., 1986). In many cases, the neurons were labeled so that their place within the cochlea could be determined, and thus their “normal” CF inferred. Inspection of the results of these studies reveals very few cases where the threshold at CF is more than 90 dB higher than normal. Assuming that 60–65 dB of this threshold elevation was due to OHC damage (which is consistent with their anatomical and physiological results), this implies that the maximum elevation due to IHC damage is 25–30 dB; above this, the IHCs become completely nonfunctional, and no neural response is measurable.
The following conclusions can be drawn from the data and arguments presented so far:
Absolute thresholds between 0 and 65 dB higher than normal (50 dB at low frequencies) can be caused by “pure” OHC dysfunction or by a mixture of OHC and IHC dysfunction.
Absolute thresholds more than 65 dB higher than normal (50 dB at low frequencies) are probably associated with a combination of OHC and IHC dysfunction (assuming that there is no retrocochlear dysfunction).
Absolute thresholds more than 90 dB higher than normal (75–80 dB at low frequencies) are likely to be associated with a dead region in the cochlea.
I turn now to consideration of cases where the absolute threshold is measured for a tone falling in a dead region. As noted earlier, in such cases, the audiometric threshold will give a misleading impression of the amount of hearing loss (Gravendeel and Plomp, 1960; Humes, 1983; Humes et al., 1984; Halpin et al., 1994). Perhaps the most marked situation where this occurs is when there is a dead region at low frequencies. Examples of etiologies that are known to be associated with low-frequency dead regions are Mondini dysplasia (Parving, 1984), vascular disruption (Muziek et al., 1987) and the advanced stages of Ménière's disease (Halpin et al., 1994).
The reason why the audiometric threshold can be misleading is illustrated schematically in Figure 1. The solid curve shows the “excitation pattern” that might be evoked by a low-frequency (250 Hz) tone in an ear with a low-frequency dead region, with normal hearing at medium and high frequencies. The dead region is indicated by the shaded area. The excitation pattern can be thought of as reflecting the amount of basilar-membrane vibration, plotted as a function of “place” or the associated CF (although “excitation” may be a misleading word here, as the basilar-membrane vibration falling in the apical region of the cochlea does not give rise to any neural excitation). The low-frequency tone is not detected via neurons innervating the apical region of the cochlea, as the IHCs in that region are dead. However, the tone will become audible when it produces sufficient excitation in the region of normal hearing. In this example, the 250-Hz tone needs to be presented at about 60 dB to produce detectable excitation just outside the edge of the dead region (CFs just above 1.3 kHz).
The solid curve shows the excitation pattern that might be evoked by a low-frequency (250 Hz) tone in an ear with a low-frequency dead region, with normal hearing at medium and high frequencies. The dead region is indicated by the shaded area. The tone level was chosen so that it was at absolute threshold; the excitation in the frequency region just above 1.3 kHz lies a little above the “threshold” excitation level indicated by the horizontal dashed line.
Gravendeel and Plomp (1960) were among the first to discuss the possibility of dead regions in people with low-frequency hearing loss, together with near-normal hearing at higher frequencies. They stated that “Patients with a bass deafness of the type described by us are actually totally deaf for low tones”. Curiously, however, they did not attribute the detection of low-frequency tones to the spread of excitation from apical regions of the cochlea to more basal regions. Instead they proposed that harmonics of the low-frequency tones were detected; these harmonics were assumed to arise in the audiometric equipment or in the ear itself. This view is almost certainly not correct. Harmonic distortion in modern equipment is usually rather low, and the ear itself does not appear to generate significant harmonic distortion (Yates, 1995). Later researchers recognized that spread of excitation was probably responsible for the detection of low-frequency tones falling in a dead region (Thornton and Abbas, 1980; Florentine and Houtsma, 1983; Turner et al., 1983; Humes et al., 1984; Halpin et al., 1994).
It is not possible to determine from the audiogram alone whether or not a patient has a low-frequency dead region. For example, Halpin et al. (1994) described two patients with very similar audiograms, both having a low-frequency hearing loss with nearly normal mid-frequency hearing. Post-mortem examination showed that one had no survival of the organ of Corti in the apical region, while the other had an organ of Corti which was present and of normal appearance. Because of the difficulty in using the audiogram to diagnose a dead region, many researchers have advocated the use of masking sounds as a way of diagnosing the presence of dead regions. The use of maskers will be described in more detail later in this paper. In the case of patients with low-frequency hearing loss, the maskers used have included highpass noise (Parving and Elberling, 1982), bandpass noise (Humes et al., 1984) or a pure tone with a frequency just inside the region of normal hearing (Thornton and Abbas, 1980; Halpin et al., 1994). The masker has been chosen to produce its greatest effect in the region of near-normal hearing. The rationale can be illustrated by considering two cases:
If there is a dead region at low frequencies, a low-frequency tone will be detected via the spread of excitation to more basal regions of the cochlea, as illustrated in Figure 1. For example, a 250-Hz tone may be detected via the spread of excitation to the 1300-Hz place. If a noise highpass filtered at 1300 Hz (or a 1300-Hz tone) is used as a masker, this will effectively mask the excitation at the 1300-Hz place. Therefore this noise (or tone) will be a very effective masker of the 250-Hz tone; the noise will produce a marked elevation in threshold.
If there is not a dead region at low frequencies, a low-frequency tone will usually be detected “locally” via IHCs and neurons with low CFs. A highpass noise (or high-frequency tone) will produce little excitation at the apical end of the cochlea (as there is usually little downward spread of excitation). Therefore, the noise will have little effect on the detection of the low-frequency tone.
To summarize: if a highpass noise (or high-frequency tone) produces a marked threshold elevation for a low-frequency signal, this indicates a low-frequency dead region. If the noise (or tone) produces little threshold elevation, this indicates the presence of functioning IHCs and neurons with low CFs. On the basis of studies of this type, it has been proposed that hearing losses of 40–50 dB at low frequencies, combined with near normal hearing at high frequencies, may be associated with a low-frequency dead region (Terkildsen, 1980; Thornton and Abbas, 1980). Humes et al. (1984) proposed that dead regions were associated with an audiogram where the loss decreased with increasing frequency at a rate exceeding 25 dB/oct. However, it should be emphasized again that it is not possible to determine from the audiogram alone whether or not a dead region is present.
In our own research (Moore et al., 2000b), which will be described in more detail later, we have found that a low-frequency dead region can be associated with a hearing loss which is relatively flat, or which decreases only slowly with increasing frequency. This can happen because cochlear hearing impairment is often associated with extensive upward spread of excitation (Leshowitz, 1977; Leshowitz and Lindstrom, 1977; Stelmachowicz et al., 1987; Murnane and Turner, 1991); the excitation produced by a low-frequency tone spreads a considerable way along the cochlea, so even when the apical region is dead, the hearing loss for a low-frequency tone is not much greater than for a mid-frequency tone. Thus, whenever a patient presents with a low-frequency hearing loss, either in the presence of near-normal hearing at middle frequencies, or when the hearing loss is relatively “flat”, a dead region should be suspected, and further tests carried out to confirm or deny this suspicion.
It is often the case that downward spread of excitation in the cochlea is very restricted (i.e. the low-frequency side of the excitation pattern is steep), even in ears where the hair cells are damaged. However, there can be considerable individual variability, and some hearing-impaired ears show marked downward spread of excitation as well as upward spread of excitation (Glasberg and Moore, 1986). Because the excitation pattern usually has a steep low-frequency side, a dead region at high frequencies is usually associated with a severe or profound hearing loss at high frequencies, and the audiogram is often steeply sloping. However, in such cases, it is unclear exactly where the dead region begins; do audiometric thresholds on the sloping part of the audiogram provide a “true” estimate of hearing loss, or do they reflect the (limited) downward spread of excitation to a region with surviving IHCs and neurons?
Recent research conducted in my laboratory suggests that a dead region can start at a frequency where the absolute threshold is near normal. Thresholds for tone frequencies falling on the steeply sloping part of the audiogram reflect the downward spread of excitation to a region of the cochlea with surviving IHCs and neurons. This is illustrated in Figure 2, which shows excitation patterns calculated for a hypothetical ear with a 40-dB hearing loss at low frequencies, and a dead region extending from 1 kHz upwards (indicated by the shaded area). The patterns were calculated using the model described by Moore and Glasberg (1997) which takes into account the broadening of the excitation patterns that occurs with increasing level and increasing OHC damage (Moore, 1998). Each of the curves represents the excitation pattern for a tone with frequency falling in the dead region; the frequencies used are 1.1, 1.2, 1.3, 1.4 and 1.5 kHz. It is assumed that this tone is detected because of the downward spread of excitation. To be detectable the excitation must fall above the horizontal dashed line, whose position is determined by the absolute threshold for frequencies below the dead region. The level of each tone has been chosen so that the excitation spreading to the region with CFs just below 1 kHz is just detectable. The solid circles represent the frequency and level at the peak of each excitation pattern. The function traced out by the circles is closely related to the expected absolute threshold for such a hearing loss, specified in dB SPL. Notice that the expected threshold changes by about 32 dB as the frequency changes from 1.1 to 1.5 kHz, which corresponds to a relatively steep slope of 72 dB/octave.
Excitation patterns calculated for a hypothetical ear with a 40-dB hearing loss at low frequencies (threshold excitation level indicated by the horizontal dashed line), and a dead region extending from 1 kHz upwards (indicated by the shaded area). Each of the curves represents the excitation pattern for a tone with frequency falling in the dead region; the frequencies used are 1.1, 1.2, 1.3, 1.4 and 1.5 kHz. The levels of the tones were chosen so that each was at the absolute threshold. Solid circles indicate the levels at the peaks of the excitation patterns.
Whenever the audiogram has a very steep slope, the threshold worsening rapidly with increasing frequency, this should be taken as preliminary evidence for a dead region, and further testing carried out. However, dead regions do sometimes occur when the audiogram is not steeply sloping (Moore et al., 2000b), so the slope of the audiogram cannot be taken as a reliable way of assessing the presence or absence of dead regions.
Mid-frequency dead regions, with good hearing at low and high frequencies, appear to be rather rare, but they do sometimes occur (Moore, 1998; Moore and Alcántara, 2001). They may be associated with a mid-frequency notch in the audiogram, or with a “cookie bite” loss. When the dead region is restricted in extent, the functional consequences seem to be slight. For example, one patient described in Moore and Alcántara (2001) was diagnosed as having a dead region extending from 1300 to 2800 Hz (he had a similar loss in the two ears). He reported no serious problems in understanding speech, and did not wear a hearing aid. Evidently, his residual hearing at low and high frequencies was sufficient to allow good speech comprehension.
General Conclusions on the Effect of Dead Regions on the Audiogram
It should be clear from the above arguments that it is difficult to determine from the audiogram alone whether or not there is a dead region. It is even more difficult to define the extent of any dead region that might be present. However, the following features in the audiogram can be taken as strong hints that a dead region may be present:
A hearing loss more than 90 dB at high frequencies or 75–80 dB at low frequencies.
A hearing loss of 40–50 dB at low frequencies with near-normal hearing at medium and high frequencies (perhaps indicating a low-frequency dead region).
A hearing loss greater than 50 dB at low frequencies with somewhat less hearing loss at higher frequencies (also perhaps indicating a low-frequency dead region).
A hearing loss increasing rapidly (more than 50 dB/octave) with increasing frequency (perhaps indicating a high-frequency dead region).
Detecting and Diagnosing Dead Regions Using Masking
As noted above, several researchers have advocated the use of masking as a way of detecting the presence of dead regions. The use of high-pass noise or a single tone as a masker has already been described. In this section, I review two other types of masking experiments that have been used to define dead regions more precisely.
Psychophysical Tuning Curves
The measurement of psychophysical tuning curves (PTCs) (Chistovich, 1957; Small, 1959) involves a procedure which is analogous to the physiological determination of a tuning curve on the basilar membrane (Sellick et al., 1982) or a neural tuning curve (Kiang et al., 1965). The signal is fixed in frequency and in level, usually at a level just above the absolute threshold, say, 10 dB Sensation Level (SL). The masker can be either a sinusoid or a narrow band of noise; often a band of noise is used to reduce the influence of beats between the signal and masker (Egan and Hake, 1950; Moore et al., 1998). For each of several masker center frequencies, the level of the masker needed just to mask the signal is determined. For normally hearing subjects, the tip of the PTC (i.e. the frequency at which the masker level is lowest) always lies close to the signal frequency (Vogten, 1974; Moore, 1978). Put another way, the masker is most effective when its frequency is close to that of the signal.
When hearing-impaired listeners are tested, PTCs have sometimes been found whose tips are shifted well away from the signal frequency (Thornton and Abbas, 1980; Florentine and Houtsma, 1983; Turner et al., 1983; Moore et al., 2000b; Moore and Alcántara, 2001). This can happen when the signal frequency falls in a dead region. The most-studied situation of this type is when there is a low-frequency dead region. The detection of low-frequency tones is then mediated by neurons with high CFs. If the signal to be detected has a frequency corresponding to the dead region, the tip of the tuning curve lies well above the signal frequency. In other words, a masker centered well above the signal in frequency is more effective than a masker centered close to the signal frequency. This happens because the higher-frequency masker lies closer to the CFs of the neurons mediating detection of the signal.
An example of a PTC for a patient with an extensive low-frequency dead region is shown in Figure 3. The masker was a narrowband noise. The patient had a relatively flat hearing loss, but her hearing improved slightly around 4 kHz. The PTC was obtained using a 2-kHz signal (indicated by the filled circle). The tip of the PTC is clearly shifted, to a frequency of 4 kHz. Sometimes, the tip of the PTC can be very substantially shifted. For example, a signal frequency of 400 Hz can be associated with a PTC with a tip at about 3500 Hz
A PTC obtained from a patient with an extensive low-frequency dead region. The patient had a relatively flat hearing loss, but her hearing improved slightly around 4 kHz. The PTC was obtained using a 2-kHz signal (indicated by the filled circle). The tip of the PTC is clearly shifted, to a frequency of 4 kHz.
When the signal frequency falls in a dead region situated at medium frequencies, the tip of the PTC may be shifted either upwards or downwards, depending on the signal frequency relative to the edges of the dead region (Moore, 1998; Moore and Alcántara, 2001). If the signal frequency is close to the low-frequency boundary of the dead region, the signal will be detected via downward spread of excitation, and the tip of the PTC will be shifted downwards. If the signal frequency is close to the high-frequency boundary of the dead region, the signal will be detected via upward spread of excitation, and the tip of the PTC will be shifted upwards. When a dead region falls at high frequencies, the signal is usually detected via the downward spread of excitation, and the tip of the PTC is shifted towards lower frequencies. An example is given in Figure 4. Again the masker was a narrowband noise.
A PTC obtained from a subject with a high-frequency dead region. The PTC was obtained using a 1.5-kHz signal (indicated by the filled circle). The tip of the PTC is shifted towards lower frequencies.
According to the arguments given above, if the signal frequency falls within a dead region, then the absolute threshold of the signal will be reached when its level is high enough to produce a detectable spread of excitation to an adjacent region where there are surviving IHCs and neurons (see Figures 1 and 2). When the signal is increased in level above the absolute threshold, the maximum neural excitation produced by the signal should always occur in IHCs and neurons with CFs just outside the edge of the dead region. Therefore, when a PTC is determined, the masker should always be most effective when its frequency lies just outside the edge of the dead region. This leads to the prediction that, when the signal frequency falls in a dead region, the tip of the PTC should fall at the boundary of the dead region, and it should not shift in frequency when the frequency or the level of the signal are changed.
Data on the effect of signal frequency are mostly in accord with this prediction. An example is given in Figure 5, which shows PTCs obtained from a subject with an extensive low-frequency dead region; the data are taken from Moore and Alcántara (2001). The masker was a narrowband noise. PTCs are shown for signal frequencies of 400, 1000 and 1500 Hz. The PTCs for the different signal frequencies overlap, especially on the low-frequency side, and all three have tips falling around 3000 Hz. These results are consistent with the idea that all three signals were detected at the 3000-Hz place, and that 3000 Hz corresponds to the upper boundary of the dead region. Similar data have been presented by Thornton and Abbas (1980).
PTCs obtained from a subject with an extensive low-frequency dead region; the data are taken from Moore and Alcántara (2001). PTCs are shown for signal frequencies of 400, 1000 and 1500 Hz, as indicated in the key.
Empirical data are not entirely in accord with the prediction that the frequency at the tip of the PTC should be independent of signal level. Figure 6 shows data obtained by Florentine and Houtsma (1983) from a subject with one normal ear and one ear with a hearing loss that was most severe at low frequencies. Note that the masker in this case was a sinusoid, which is generally a less effective masker than a narrowband noise; hence the masker level required to mask the signal has to be much higher than would be the case with a narrowband noise masker. For the normal ear, a PTC was measured using a 1-kHz signal at 7 dB SL. The PTC, shown in the left part of the figure, has a tip at the signal frequency (although data were not obtained when the masker frequency was very close to the signal frequency, to reduce problems with beat detection). The impaired ear was tested using a 1-kHz signal at 4.5, 7 or 13 dB SL. The resulting PTCs are shown by the triangles, circles and squares in the right part of the figure. All three PTCs show tips that lie well above the signal frequency, which is consistent with a low-frequency dead region. However, the tips clearly shift downwards in frequency as the signal level increases, which is not consistent with the prediction given above. In our own work (Moore and Alcántara, 2001) we have found some small shifts in the frequencies at the tips of the PTCs when the signal level is altered, but the shifts have not been as large as found by Florentine and Houtsma.
Data obtained by Florentine and Houtsma (1983) from a subject with one normal ear and one ear with a hearing loss that was most severe at low frequencies. For the normal ear, a PTC was measured using a 1-kHz signal at 7 dB SL (left). The impaired ear was tested using a 1-kHz signal at 4.5, 7 or 13 dB SL. The resulting PTCs are shown by the triangles, circles and squares in the right part of the figure. Crosses indicate absolute thresholds.
Florentine and Houtsma (1983) suggested two explanations for the shift in frequency of the PTC tips with signal level. The first was that “the shift may be caused by fibers that respond to high-level stimulation although they are damaged”. Re-phrased in the terminology of this paper, there may have been IHCs with CFs somewhat above the signal frequency, which were damaged but not dead. For example, IHCs tuned around 2.2 kHz may have fallen into this category; perhaps they did not respond to a 1-kHz tone at 4.5 dB SL (about 71 dB SPL), but they did respond to a 1-kHz tone at 13 dB SL (about 79 dB SPL). The second explanation suggested by Florentine and Houtsma was that the shift “may be caused by high-threshold fibers that are still intact”. Functionally, this argument is similar to their first argument.
Another factor that could contribute to the shift is the effect of overall level on the traveling wave on the basilar membrane. There is evidence that the peak of the traveling wave shifts in a basal direction with increasing level (McFadden, 1986; Ruggero et al., 1997). This means that, for a fixed place on the basilar membrane, the CF shifts downwards with increasing level. A dead region has a fixed location on the basilar membrane. Thus, if there is a low-frequency dead region, the CFs of the surviving IHCs and neurons at the boundary of the dead region will shift downwards with increasing level. When a PTC is measured using a signal frequency within the dead region, the masker should be most effective when its frequency matches the CF of the surviving IHCs at the boundary of the dead region. Since this CF decreases with increasing level, the frequency at the tip of the PTC should decrease with increasing signal level, as observed by Florentine and Houtsma.
A weakness of this explanation is that level-dependent shifts of the traveling wave appear to occur only in ears where the active mechanism is operating (i.e. where the OHCs are still functioning). In a hearing-impaired ear, the shifts ought to be small. However, the subject tested by Florentine and Houtsma had only a mild hearing loss for frequencies of 2.5 kHz and above, so some shift in the traveling wave with level may have occurred.
Whatever the explanation for the shifts in the PTC tips observed by Florentine and Houtsma (1983), it is noteworthy that the tips of the PTCs did not shift back to the signal frequency, even for the highest signal level used. In my laboratory, we have never found a case where the tip of the PTC was shifted markedly in frequency for low signal levels, but shifted to the signal frequency for higher levels, even when the signal level was as high as 21 dB SL. It seems likely therefore, that when the tip of the PTC is markedly shifted, there is a “true” dead region around the signal frequency, where the IHCs are completely non-functional. The situation may be different when the tip of the PTC is only slightly shifted, There may be a “transition” region between a dead region and a region with functioning IHCs where the IHCs are damaged, but capable of responding at sufficiently high levels. If the signal frequency falls in this transition region, the tip of the PTC may be shifted away from the signal frequency for low signal levels, but may move towards the signal frequency at higher signal levels. Further experimental work is needed to assess how often this occurs.
One complication in using PTCs to estimate the boundary of a dead region occurs when the signal frequency lies very close to the boundary, but still inside the dead region. In such a case, one would expect the tip of the PTC to be shifted only slightly from the signal frequency. When the masker and signal frequencies of the signal are similar, beats occur between the signal and masker. The beats are most clearly audible when the masker is a sinusoid, but they can play some role even when a narrowband noise masker is used (Moore et al., 1998; Derleth et al., 1999; Alcántara et al., 2000; Moore and Alcántara, 2001). The beats provide a cue for the detection of the signal, and the salience of the beats (and hence their effectiveness as a cue) depends on the beat rate, which is equal to the masker-signal frequency separation; the beats are most salient at low rates, and become inaudible for high rates (above about 700 Hz; see Alcántara et al., 2000). Beats are not audible when the masker and signal frequencies are equal. The variation in the availability and salience of detection cues with masker-signal frequency separation can result in a “distortion” of the shape of the PTC around its tip (Moore and Alcántara, 2001). Thus, when the signal frequency lies within a few hundred Hertz of the boundary of a dead region, the tip of the PTC may not indicate accurately where the boundary lies. This effect is more important when the boundary lies at a low frequency than when it lies at a high frequency, as the error in estimating the boundary frequency is roughly constant in Hertz, so the ratio of the estimated to the “true” boundary frequency becomes closer to one with increasing frequency.
In summary, PTCs provide a useful way of detecting dead regions and defining their boundaries. When the tip of the PTC is shifted markedly from the signal frequency, then, to a first approximation, the frequency at the tip indicates the boundary of the dead region. However, the frequency at the tip of the PTC may not give an accurate estimate of the boundary when the shift is small (less than a few hundred Hertz).
Although PTCs have been very useful in laboratory studies of dead regions, they are time-consuming to measure, and the choice of an appropriate signal frequency and level can be difficult. I turn now to a simpler method of diagnosing dead regions, based on the use of a broadband noise masker.
Introduction
I now describe a test that is simple enough to be used clinically to detect the presence of one or more dead regions, and to delimit the frequency range of any dead region. The test is based upon the detection of sinusoids in the presence of a broadband noise, designed to produce almost equal masked thresholds (in dB SPL) over a wide frequency range, for normally hearing listeners and for listeners with hearing impairment but without dead regions. This noise is called threshold equalizing noise (TEN) (Moore et al, 2000b). An abnormally high masked threshold at a particular frequency is taken to indicate a dead region at that frequency, for reasons discussed below. I should emphasize that the rationale behind the test does not require the use of TEN; in principle other types of broadband noise could also be used. However, the test is particularly simple to apply when TEN is used, because it is then easy to identify deviations from “normal” masked thresholds. To increase simplicity even more, the noise level is calibrated in such a way (described below) that the “normal” masked threshold for any frequency within the range 250 to 10000 Hz is very close to the nominal noise level. For example, a nominal noise level of 70 dB should give a “normal” masked threshold of 70 dB SPL.
It should be emphasized that TEN is not the same as uniformly masking noise (Zwicker and Fastl, 1990); the latter is intended to produce an equal amount of masking at all frequencies (i.e. to raise thresholds a constant amount above the absolute threshold), while the former is intended to give constant masked thresholds in dB SPL. TEN is also not the same as uniformly exciting noise (Glasberg and Moore, 2000); the latter is intended to produce a constant amount of excitation at each CF, and takes into account the effects of transmission of the noise through the outer and middle ear. The TEN does not produce equal excitation at all CFs, although it does this approximately for mid-range frequencies, from about 500 to 5000 Hz.
Langenbeck (1965) was an early proponent of what he called “above-threshold audiometry” or “noise audiometry”. He thought that the results of the test could be used to distinguish auditory nerve damage from hair cell damage. He argued that if the threshold for detecting a tone in the noise was much higher than normal, and was also above the absolute threshold, then this was indicative of nerve damage. He stated that “In ganglionar or acoustic nerve lesions the tones in the region of the lesion can be masked with abnormal ease” (page 133). On the other hand, if the masked threshold was close to the normal value, this was thought to indicate inner hair cell damage. At the time he did his work, the differences in function of the OHCs and IHCs were not appreciated. Langenbeck does not appear to have realized the possibility that a tone of a given frequency might be detected via the spread of excitation to neurons with adjacent CFs. Indeed, he specifically talked about “tonal stimulation at one point” and assumed that the threshold corresponded to a certain change in neural activity at that point. He assumed that higher masked thresholds were associated with neural damage because, when the tone is added to the noise “the chances of releasing an additional nerve impulse are smaller, because of the loss of fibres” (page 136).
Our test is similar to that of Langenbeck, and the Langenbeck test has been used previously in the diagnosis of dead regions (Gravendeel and Plomp, 1960). However, our rationale is different from Langenbeck's, as will be explained below.
Characteristics of the TEN
To determine the required spectral characteristics of the TEN, we assumed that the power of the signal at threshold, Ps, is given by the equation:
where No is the noise power spectral density, K is the signal-to-noise ratio at the output of the auditory filter required for threshold and ERB is the equivalent rectangular bandwidth of the auditory filter (Patterson and Moore, 1986). The TEN was spectrally shaped so that No.K.ERB was constant over the frequency range from 125 to 15000 Hz. Values of K and the ERB were taken from Moore et al. (1997) (the variation of the ERB with level was ignored). In what follows, the noise level will be specified in terms of the level in a one-ERB wide band around 1000 Hz (i.e. the level in the frequency range 935 to 1065 Hz). An example of the noise spectrum is shown in Figure 7, for a level/ERB of 70 dB.
Spectrum of the TEN for a level/ERB of 70 dB.
To make the test easy to use, and widely available, the TEN was recorded on one channel of a CD. A sinusoidal test signal was simultaneously digitally generated and recorded on the other channel of the CD. Several different test frequencies were used, and each was assigned a unique track number on the CD. Test frequencies were 250, 500, 1000, 1500, 2000, 3000, 4000, 5000, 6000, 8000 and 10000 Hz.
To check that the TEN did indeed give equal masked thresholds at all frequencies for normally hearing subjects, masked thresholds in the TEN were measured for ten normally hearing subjects (Moore et al., 2000b). All of these subjects had absolute thresholds better than 15 dB HL at all audiometric frequencies, and they had no history of hearing disorders. Only one ear of each subject was tested. Thresholds were measured using manual audiometry, with the procedure proposed by Carhart and Jerger (1959). This method was chosen since we wished the test to be easy to use in clinical practice. The TEN from the CD was fed to one of the tape inputs on a Grason-Stadler GSI-16 audiometer, and the sinusoidal test signal was fed to the other. TEN and signal levels were controlled using the level controls on the audiometer. The noise and sinusoidal signal were mixed using the audiometer and stimuli were delivered using the Telephonics TDH50 earphones supplied with the audiometer. Each ear of each subject was tested separately.
The results are shown in Figure 8. Masked thresholds are specified in dB SPL, not dB HL. The mean thresholds for a given noise level are almost invariant across frequency, and are approximately equal to the TEN level/ERB. For example, a TEN level of 70 dB/ERB led to masked thresholds of roughly 70 dB SPL, regardless of signal frequency. The only exceptions occurred for the lowest TEN level used, where the mean threshold was slightly above 30 dB SPL at 125 Hz and markedly above 30 dB SPL at 10000 Hz. These are both cases where the masked threshold was close to the absolute threshold for some subjects. Thus, provided the TEN produces sufficient masking, it lives up to its name; it produces nearly equal masked thresholds for all signal frequencies. The error bars indicate 95% confidence intervals. On the upper side, the limits of the confidence intervals fall within 7 dB of the TEN level/ERB for frequencies up to 6000 Hz, and (for the highest TEN level) within 10 dB for frequencies up to 10 kHz. Thus masked thresholds falling 10 dB or more above the TEN level/ERB can be regarded as falling outside the normal range.
Masked thresholds in the TEN, specified in dB SPL, for TEN levels of 30, 50 and 70 dB/ERB, averaged across ten normally hearing subjects. Error bars indicate 95% confidence intervals.
In specifying the masked threshold in dB SPL, no allowance was made for the non-flat frequency response of the earphone (the response of the TDH50 earphone at the eardrum is relatively flat over the range 500 to 5000 Hz, but rolls off outside this range). The specified levels are accurate at 1 kHz, but there were deviations from the specified values at other frequencies. It seems reasonable to assume that minor irregularities in the frequency response of the earphone should have only a small effect on the outcome of the test. For example, if the response of the earphone shows a peak at, say, 1500 Hz, that peak will boost the effective level both of a 1500-Hz signal and of the noise components around 1500 Hz. The peak will not alter the local signal-to-noise ratio, and hence should not alter the masked threshold. Of course, this will not be the case if the response of the earphone changes rapidly with frequency.
For listeners with cochlear hearing loss, but without a dead region at the signal frequency, masked thresholds in the TEN are expected to be only slightly higher than for normally hearing listeners. Damage to the OHCs is associated with loss of frequency selectivity (broadening of the auditory filter) (Pick et al., 1977; Glasberg and Moore, 1986; Tyler, 1986; Moore, 1998). Typically, the ERB of the auditory filter is 2–3 times greater than the average value for normally hearing listeners, and it appears to be never more than about 3.8 times greater than normal (Moore and Glasberg, 1997). If the ERB is, say, 3 times greater than normal, then, for a given noise spectrum level, the noise power passing through the filter is 4.8 dB higher than normal (10log103 = 4.8). If threshold corresponded to a constant signal-to-masker ratio at the output of the auditory filter (Fletcher, 1940), this would lead to a threshold that was 4.8 dB higher than normal. In practice, the signal-to-noise ratio at the output of the auditory filter required to reach threshold is often 2–3 dB lower for hearing-impaired subjects than for normally hearing subjects (Glasberg and Moore, 1986), possibly because the inherent variability in the noise at the output of the filter is reduced by its greater effective bandwidth (Patterson and Henning, 1977). Thus, we would expect that, on average, masked thresholds in the TEN for hearing-impaired listeners without dead regions would be only 2–3 dB higher than normal. The results obtained by Moore et al. (2000b) support this expectation.
Expected Results Using the TEN with Hearing-Impaired Subjects with Dead Regions
Assume that a tone with frequency falling in a dead region is detected using neurons with CFs remote from the signal frequency (hereafter called “off-frequency listening”). The amplitude of basilar-membrane vibration at the remote place will generally be less than the amplitude in the dead region. Therefore, a broadband noise may mask that tone much more effectively than would normally be the case, as the noise only has to mask the reduced response at the remote place. Thus, if the threshold for detecting a tone in broadband noise is markedly higher than normal, this indicates a lack of surviving IHCs/neurons with CFs corresponding to the frequency of the tone, i.e. a dead region. The definition of “markedly” will be made more precise later.
An illustration of this rationale for a hypothetical patient with a high-frequency dead region is given in Figure 9. Consider first the top panel. It is assumed that the dead region starts at about 1.07 kHz (indicated by the shaded area), and that the excitation level required to reach absolute threshold for frequencies below 1.07 kHz is 40 dB (indicated by the horizontal long-dashed line). The solid curve shows the excitation pattern for a tone with a frequency of 1.5 kHz, which falls within the dead region. The level of the tone has been chosen so that the excitation immediately adjacent to the dead region just reaches the threshold value. In other words the 1.5-kHz tone is at the level required to reach absolute threshold. This level is about 67 dB. Consider now the bottom panel. This shows the effect of adding a broadband noise which produces an equal excitation level of 70 dB at all frequencies (the TEN does not quite do this, as noted above, but a constant excitation level is shown here for simplicity). The 1.5-kHz tone at 67 dB would be completely masked by this noise, as the noise would “swamp” the previously audible excitation at CFs just below 1.07 kHz. To restore audibility of the tone in the noise, its level has to be increased to the point where the tone-evoked excitation at CFs just below 1.07 kHz just exceeds 70 dB. The solid curve shows the excitation pattern for a tone level of 97 dB, which is the level required to achieve this. Thus, the masked threshold of the tone is 30 dB higher than the absolute threshold, and the masked threshold is 27 dB higher than the “normal” masked threshold, which would be around 70 dB.
Illustration of the rationale behind the test using TEN for a hypothetical patient with a dead region above 1.07 kHz.
An illustration of this rationale for a hypothetical patient with a low-frequency dead region is given in Figure 10. Consider first the top panel. It is assumed that the dead region extends up to 2.53 kHz (indicated by the shaded area), and that the excitation level required to reach absolute threshold for frequencies above 2.53 kHz is 40 dB (indicated by the horizontal long-dashed line). The solid curve shows the excitation pattern for a tone with a frequency of 1.5 kHz, which falls within the dead region. The level of the tone has been chosen so that the excitation immediately adjacent to the dead region just reaches the threshold value; the tone is at the absolute threshold of about 67 dB. Consider now the bottom panel. This shows the effect of adding a broadband noise producing an excitation level of 70 dB at all frequencies. The 1.5-kHz tone at 67 dB is masked by the noise. To restore audibility of the tone in the noise, its level has to be increased to the point where the tone-evoked excitation at CFs just above 2.53 kHz just exceeds 70 dB. The solid curve shows the excitation pattern for a tone level of 88 dB, which is the level required to achieve this. Thus, the masked threshold of the tone is 21 dB higher than the absolute threshold, and the masked threshold is 18 dB higher than the normal masked threshold of 70 dB.
Illustration of the rationale behind the test using TEN for a hypothetical patient with a dead region below 2.53 kHz.
To assess the validity of the TEN test, PTCs were measured using the same hearing-impaired listeners as tested with the TEN. The specific hypothesis tested was that higher-than-normal thresholds in the TEN would be associated with PTCs with shifted tips. Results consistent with this hypothesis would confirm the validity of the TEN test as a method for diagnosing dead regions. Full results of this validation are presented in Moore et al. (2000b). I describe here some illustrative cases.
Results for a hearing-impaired person who does not appear to have a dead region are shown in Figure 11. The lower panel shows results obtained with the TEN, except that filled squares indicate absolute thresholds (obtained using the same audiometer and earphones as employed for measuring thresholds in the TEN, and specified in dB SPL, but uncorrected for the non-flat frequency response of the earphone). This person has near-normal hearing for frequencies up to 1500 Hz, but a moderate-to-severe loss at higher frequencies. Over the frequency range where the TEN produces masking, the masked thresholds are only slightly higher than normal, being around 71 to 75 dB for the TEN level of 70 dB/ERB. For frequencies of 3000 Hz and above, the TEN level of 70 dB/ERB is not sufficient to produce masking, so the masked thresholds are close to the absolute thresholds.
Results for a hearing-impaired person who does not have a dead region. The lower panel shows results obtained with the TEN, except that filled squares indicate absolute thresholds. The upper panel shows PTCs determined for three signal frequencies. In each case, the signal level and frequency are indicated by a filled symbol. The corresponding PTCs are indicated by open symbols of the same shape.
The upper panel of Figure 11 shows PTCs determined for three signal frequencies. In each case, the signal level and frequency are indicated by a filled symbol. The corresponding PTC is indicated by an open symbol of the same shape. For each PTC, the tip is close to the signal frequency, although the tip is quite broad for the 4000-Hz signal. The PTCs are consistent with the results using the TEN, indicating that each signal was detected via IHCs/neurons with CFs close to the signal frequency.
Figure 12 shows an example of results for a hearing-impaired person who probably does have a dead region. This person has near-normal hearing for frequencies up to 1000 Hz, but a severe-to-profound loss at higher frequencies. In this figure, the symbols with up-pointing arrows indicate cases where the threshold was too high to be measured; the highest measurable threshold was 120 dB SPL. The specific symbol used with the arrow indicates the lowest TEN level for which a threshold could not be measured. For example, for a signal frequency of 3000 Hz, a masked threshold could be measured for the TEN level of 30 dB/ERB, but not for levels of 50 and 70 dB/ERB, so the symbol associated with the arrow at 3000 Hz is a diamond, the symbol for 50 dB/ERB. A filled square with an arrow indicates that the absolute threshold was too high to be measured.
Results for a person with a dead region at high frequencies. Otherwise, as Figure 11. Symbols with up-pointing arrows indicate cases where the threshold was too high to be measured; the highest measurable threshold (determined by equipment limitations) was 120 dB SPL. The specific symbol used with the arrow indicates the lowest TEN level for which a threshold could not be measured.
For signal frequencies of 1500 Hz and above, masked thresholds in the 70 dB/ERB noise were 10 dB or more higher than the mean normal value. For signal frequencies from 3000 to 5000 Hz, the masked thresholds in the 30 dB/ERB noise were elevated above the absolute thresholds and were at 120 dB SPL, i.e. 90 dB higher than for normal-hearing subjects! This strongly suggests that tones with frequencies of 1500 Hz and above were being detected via IHCs/neurons with CFs below 1500 Hz.
The PTC for this subject for a signal frequency of 500 Hz has a tip at 500 Hz. However, the PTCs for signal frequencies of 1200 and 1500 Hz are shifted downwards to about 1000–1200 Hz. This suggests that the dead region starts at 1000–1200 Hz, and extends upwards from there, which is consistent with the finding that thresholds in the TEN were near normal for frequencies up to 1000 Hz, but were higher than normal for frequencies of 1500 Hz and above.
In total, Moore et al. tested 20 ears of 14 subjects with sensorineural hearing loss. Generally, there was a very good correspondence between the results obtained using the TEN and the PTCs; if, for a given signal frequency, the masked threshold in the TEN was 10 dB or more higher than normal and the TEN produced at least 10-dB of masking (i.e. the masked threshold was 10 dB or more above the absolute threshold), then the tip of the PTC determined using that signal frequency was shifted. If the masked threshold in the TEN was not 10 dB or more higher than normal, the tip of the PTC was not shifted. Hence, the following “rule” was formulated: If the threshold in the TEN is 10 dB or more above the TEN level/ERB, and the TEN produces at least 10 dB of masking, this is indicative of a dead region at the signal frequency. However, some exceptions to this rule, and some limitations of the TEN test should be noted:
For some hearing-impaired subjects, the absolute thresholds at high frequencies may be so high that they cannot be measured. In such cases it is not possible to determine whether the TEN produces masking at these frequencies. In all probability, such extreme losses are associated with dead regions, although it is difficult to demonstrate this directly. However, the thresholds in the TEN for frequencies where the hearing loss is not so extreme can be used to determine whether there is a dead region corresponding to these lower frequencies and, if so, to estimate the frequency at which the dead region starts.
Sometimes, when the absolute threshold is high but still measurable, a dead region may be present (as indicated by a PTC with a shifted tip), but the TEN may not be sufficiently intense to produce 10 dB or more of masking. As a general rule, the TEN level/ERB should be chosen to be high enough to produce some masking at all frequencies of interest; for the purpose of clinical testing, frequencies up to about 5000 Hz are of the greatest interest. However, in some cases it may be impossible to make the TEN sufficiently intense, either because of limitations in the equipment, or because the TEN becomes uncomfortably loud. For most patients with moderate to severe hearing loss, a level of 70 dB/ERB is sufficient, but it may sometimes be necessary to use a TEN level of 80 or even 90 dB/ERB.
When the signal frequency falls just inside a dead region, the threshold in the TEN may be less than 10 dB higher than the TEN level per ERB. This is especially true when there is a low-frequency dead region, as illustrated in Figure 13. Consider first the top panel. It is assumed that the dead region extends up to 1.8 kHz (indicated by the shaded area), and that the excitation level required to reach absolute threshold for frequencies above 1.8 kHz is 40 dB (indicated by the horizontal long-dashed line). The solid curve shows the excitation pattern for a tone with a frequency of 1.5 kHz, which falls just inside the dead region. The tone is at the absolute threshold level of about 52 dB. The bottom panel shows the effect of adding a TEN with a level of 70 dB/ERB. To restore audibility of the tone in the noise, its level has to be increased to the point where the tone-evoked excitation at CFs just above 1.8 kHz just exceeds 70 dB. The solid curve shows the excitation pattern for a tone level of 78 dB, which is the level required to achieve this. Thus, the masked threshold of the tone is only 8 dB higher than the TEN level per ERB.
Illustration of the difficulty in applying the test TEN test when the signal frequency falls just inside a low-frequency dead region. In this example, the tone had a frequency of 1.5 kHz, and the dead region was assumed to extend up to 1.8 kHz.
Some subjects show higher than normal thresholds at all frequencies. This may indicate that they are “inefficient” or “cautious” listeners (Patterson and Moore, 1986); they may need to hear the tone very clearly before indicating that they detect it at all. Alternatively, the high thresholds may be indicative of a problem in the central auditory system (Langenbeck, 1965). Such cases need to be treated with caution. A signal threshold that is 10 dB or more higher than normal may not indicate a dead region. However, we have not so far found a case where signal thresholds are 10 dB higher than normal at all frequencies. A reasonable policy in cases where thresholds are in the range 5–10 dB higher than normal, even at frequencies where the hearing loss is mild or moderate, is to adopt a more stringent criterion for diagnosis of a dead region, for example, requiring thresholds in the TEN to be 15 or more dB higher than normal.
In spite of these exceptions to the rule, the measurement of signal thresholds in the TEN does seem to provide a reasonably reliable method for diagnosing the presence of dead regions and defining their limits. “False positives”—the diagnosis of a dead region at a specific frequency when there is not a dead region—appear to be rare. The test is rather quick to apply. Assuming that absolute thresholds as well as masked thresholds are measured using the test stimuli recorded on the CD (the recommended procedure), the test takes approximately twice as long as conventional audiometry, if a single noise level is used.
Earlier in this paper, I discussed the possibility of cases where the IHCs are severely damaged, and hence require a larger input than normal to function, but which nevertheless are capable of initiating action potentials in the auditory nerve if the input level is sufficiently high. For signal frequencies corresponding to the CFs of such IHCs, we would expect signal thresholds measured in TEN to be markedly higher than normal when the noise produces a small amount of masking; in this case the signal would be detected via off-frequency listening. However, for higher TEN levels, the masked thresholds should approach normal values, as the on-frequency IHCs should start responding. We have never observed such a pattern of results. If masked thresholds in the TEN are abnormally high for a given TEN level, they remain abnormally high (and the signal-to-noise ratio at threshold remains about the same) when the TEN level is increased.
We can interpret this finding in the following way. If the IHCs at a given CF are so severely damaged that a low SL signal at that CF is more effectively detected via off-frequency listening than via on-frequency listening, then these IHCs do not convey useful information for the detection of a signal in noise even at higher levels. In other words, the IHCs may respond weakly at higher levels, but the signal from them is not sufficient to make on-frequency listening more effective than off-frequency listening. Functionally, the IHCs are dead, even though it might, in theory, be possible for them to lead to the generation of action potentials in the auditory nerve.
Test Procedure for Clinical Use
For routine use of the test CD in the clinic, the following procedure is recommended.
The output from the CD player should be fed to the left and right line-level inputs on the audiometer (often used for input from tape). The first track contains a calibration tone. The VU meters for the left and right channels on the audiometer should both be adjusted to read −6 dB (i.e. 6 dB below the 0 level) when this track is played. For subsequent tracks, the level per ERB (dB SPL) of the noise on the right channel is 10 dB lower than the nominal level indicated on the audiometer. The level of the sinusoid on the left channel is 10 dB higher than the indicated level. The controls on the audiometer can be used to adjust the levels of the tone and noise to the desired values.
Measure absolute thresholds in dB SPL using the test tones recorded on the left channel of the CD. This should be done separately for each ear, with masking noise in the opposite ear when the opposite ear has markedly better hearing than the test ear. Contralateral noise should also be used in such cases when masked thresholds in the TEN are being measured (step 3, below). It is unlikely that the contralateral noise will produce any central masking of the test tone, since contralateral masking produced by continuous noise is very small (Zwislocki, 1971).
Measure masked thresholds in the TEN for the same signal frequencies. The two channels of the audiometer need to be mixed in order to do this. Usually, it is only necessary to test using one level/ERB. The level/ERB to be used depends on the severity of the hearing loss of the person being tested. Ideally, the level/ERB should be higher than the lowest threshold that is measured at any frequency within the frequency range up to about 5000 Hz. Usually, a level of 70 dB/ERB will be suitable. However, if time is available, measurements at more than one level/ERB can be informative.
Dead regions for a particular frequency are indicated by a masked threshold that is at least 10 dB above the absolute threshold and 10 dB above the level/ERB of the TEN.
If a person has masked thresholds 10 dB or more above the level/ERB of the TEN even at frequencies where the hearing loss is mild or moderate, caution is needed in interpreting the results; the high thresholds may be the result of retrocochlear problems rather than dead regions.
If there is any ambiguity, for example, if the masked threshold is higher than normal, but is only 5 dB above the absolute threshold, then, when possible, the test should be repeated with a higher noise level. It may not be possible to use levels higher than 80 dB/ERB, as the overall noise level then becomes rather high, and it may be uncomfortably loud for some patients.
The Perception of Speech by People with Dead Regions
Theoretical Background
There are several theoretical reasons why people with dead regions might extract little or no information from frequency components of speech that fall within a dead region, even when those components are amplified sufficiently to make them audible. These reasons include:
The frequency components are received through the “wrong” place in the cochlea. For example, if there is a low-frequency dead region, amplified low-frequency components will be detected and analyzed via the frequency channels that are tuned to higher frequencies. This mismatch between frequency and place may lead to difficulty in interpreting the information derived from the low frequencies. There is some evidence supporting this idea from studies involving the simulation of hearing loss and/or of cochlear implant signal processing (Shannon et al., 1998). However, extended learning with “remapped” stimuli may partially compensate for this problem (Rosen et al., 1999).
If the components falling in the dead region are amplified sufficiently to make them audible, they will be detected and analyzed via the same neural channels that are used for other frequencies, and this may impair the analysis of those other frequencies. For example, if there is a low-frequency dead region, the amplified low-frequency components will be detected and analyzed through the same neural channels as are used for the medium/high frequencies. Since speech is a broadband signal, usually containing components covering a wide frequency range, this may lead to some form of “information overload” in those channels.
Information in speech, such as information about formant frequencies, may partly be coded in the time patterns of the neural impulses (phase locking). The analysis of temporal information may normally be done on a place-specific basis. For example, the neural machinery required to “decode” temporal information about frequencies around 1000 Hz may be restricted to neural channels with CFs close to 1000 Hz (Loeb et al., 1983; Srulovicz and Goldstein, 1983). This is the theoretical rationale behind the measure “average localized synchronized rate” used by Sachs, Young and co-workers (Sachs and Young, 1980; Miller et al., 1997). When there is a mismatch between the frequencies of the speech components and the place where they are detected, the temporal decoding mechanisms required to analyze those speech components may not operate effectively.
The relative importance of these three factors is not known; all may be important to some extent. Nevertheless, the empirical evidence is reasonably clear; only limited information can be extracted from frequency components falling in a dead region. This evidence is reviewed in the following sections.
Perception of Speech by People with Low-Frequency Dead Regions
Thornton and Abbas (1980) examined the perception of speech using three subjects who had been diagnosed as having low-frequency dead regions on the basis of PTCs with upwards-shifted tips and on the basis of masking effects produced by a high-frequency tone. The speech materials were word lists that were filtered in various ways. For the broadband speech (unfiltered, presented at 80 dB HL) scores ranged from 56 to 88%, while scores for a control group of five normally hearing subjects ranged from 94 to 100%. When the speech was high-pass filtered at 3000 Hz, scores worsened, and were similar for the two groups (34 to 46% for the subjects with dead regions and 16 to 38% for the control group). Thus, the subjects with low-frequency dead regions were able use the high-frequency information in the speech as well as or better than the control group. When the speech was low-pass filtered at 1500 Hz, performance worsened dramatically for the group with dead regions; scores dropped to 12 to 44%. In contrast, scores for the control group remained rather high, at 76 to 84%. Finally, when the speech was low-pass filtered at 1500 Hz and presented simultaneously with noise that was highpass filtered at 1500 Hz, scores dropped even further for the group with dead regions (0 to 22%), but remained high for the control group (64 to 80%). The effect of the noise suggests that, for the group with dead regions, low-frequency speech information may have been partly obtained via IHCs and neurons tuned to middle or high frequencies. Overall, the results suggest that the group with dead regions made little use of low-frequency information in the speech, but they could use high-frequency information efficiently.
A case study of a person with a low-frequency dead region was presented by van Tasell and Turner (1984). The dead region had been diagnosed on the basis of PTCs with upward-shifted tips (Turner et al., 1983). The speech stimuli used were monosyllabic words, sentences, and nonsense syllables. A control group of normally hearing subjects was also tested. Speech recognition scores of all subjects were close to 100% correct for all stimuli presented unfiltered at a moderate intensity level. When stimuli were low-pass filtered, the performance of the hearing-impaired subject fell below that of the control group, but remained considerably above chance. A further diminution in the impaired subject's recognition of nonsense syllables resulted from the addition of a high-pass masking noise, indicating that his performance in the filtered quiet condition was attributable in large part to information carried via IHCs and neurons with medium to high CFs. Scores for the control group were also somewhat decreased by the masker, suggesting that they also may have been extracting some low-frequency speech cues from middle to basal regions of the cochlea.
Halpin et al. (1994) measured the recognition of words in fourteen patients with low-frequency hearing loss who were divided into two groups: seven were diagnosed as having “true” hearing at low frequencies, and the other seven were diagnosed as having low-frequency dead regions using a test based on the masking of low-frequency tones by a high-frequency tone (at about 85 dB HL) situated at the border of the region of hearing loss. Speech recognition scores were generally lower (10 to 88%) for the group with dead regions than for the group without dead regions (84 to 100%). Obtained scores were compared to scores predicted from the articulation index (AI; see ANSI, 1969). The predictions were based either on the “true” audiograms, or on “hypothetical” audiograms where the hearing loss at low frequencies was set to a large value to simulate low-frequency dead regions. For the patients without dead regions, the best correspondence between obtained and predicted scores was obtained using the “true” audiograms for the predictions. For the patients with dead regions, the best correspondence was obtained using the “hypothetical” audiograms.
Halpin et al. concluded that, for broadband speech, patients with dead regions extracted little or no information from low-frequency components in the speech. They argued further that, for patients with low-frequency hearing loss, amplification of the low frequencies via a hearing aid should only be provided when there was not a dead region at low frequencies, since “patients with nonfunctional apices may be unusually sensitive to the costs imposed by amplification”.
The studies reviewed above are all consistent with the idea that, for patients with low-frequency dead regions, only limited information can be extracted from low-frequencies in the speech. However, there are at least hints that some small amount of low-frequency information can be obtained via IHCs and neurons tuned to medium and high frequencies.
Perception of Speech by People with High-Frequency Dead Regions
There have been reports over a period of many years suggesting that people with moderate-to-severe hearing loss at high frequencies often do not benefit from amplification of high frequencies, or even perform more poorly when high frequencies are amplified (Villchur, 1973; Moore et al., 1985; Murray and Byrne, 1986; Ching et al., 1998; Hogan and Turner, 1998; Turner and Cummings, 1999; Amos and Humes, 2000). Unfortunately, in the great majority of studies, it is not clear whether or not a high-frequency dead region was present in any specific patient or group of patients.
Murray and Byrne (1986) tested five subjects with near-normal hearing for frequencies up to 1 kHz, and losses of 65 to 80 dB between 4 and 8 kHz. Speech stimuli were initially amplified using the frequency-gain characteristic prescribed by the NAL(R) procedure (Byrne and Dillon, 1986); this was done separately for each subject. Then the stimuli were lowpass filtered with cutoff frequencies of 4.5, 3.5, 2.5 or 1.5 kHz. They presented running (continuous) speech in speech-shaped noise. The (input) noise level was fixed at 70 dBA, and subjects were asked to adjust the speech level until 50% of the speech could be understood. This speech level, expressed as a speech-to-noise ratio, was referred to as the speech reception threshold (SRT). Three of the subjects performed better (the SRT was lower) when the cutoff frequency was 2.5 or 3.5 kHz than when it was 4.5 kHz. For the other two subjects, there was little change in SRT as the cutoff frequency was decreased from 4.5 to 2.5 kHz. For four of the five subjects, performance worsened when the cutoff frequency was decreased from 2.5 to 1.5 kHz.
These results indicate that, for people with moderate-to-severe high-frequency hearing loss, amplification of the high frequencies in speech is not always beneficial, and may sometimes be harmful. However, it is noteworthy that amplification of frequencies between 1.5 and 2.5 kHz was beneficial for four of the five subjects. From the shapes of their audiograms, it might be suspected that these subjects had high-frequency dead regions starting at about 1.5 kHz. If this was the case, then the results suggest that there may be some benefit from amplification of frequencies falling a little inside a dead region. I will return to this point later.
Hogan and Turner (1998) tested nine subjects with near-normal hearing at low frequencies and high-frequency hearing losses ranging from about 40 to 110 dB. Subjects were required to identify nonsense syllables using speech that had been given high-frequency emphasis (to restore audibility) and was then lowpass filtered with various cutoff frequencies. When the high-frequency loss was greater than 55 dB, increasing the high-frequency cutoff often was of little benefit, or made performance worse. Calculations of the audibility of speech using both the “old” articulation index (ANSI, 1969) and the newer speech intelligibility index (SII) (ANSI, 1997) indicated that the relatively poor use of high-frequency information could not be explained solely in terms of audibility or listening level. Hogan and Turner calculated a measure that they called “efficiency”. This measure quantifies how well a hearing-impaired listener uses audible (amplified) speech compared to a normally hearing listener receiving speech at normal conversational levels. The measure was calculated separately for several one-third octave frequency bands. A value of 1 indicates use of speech information the same as for a normally hearing listener, a value of 0 indicates no use from that specific band, and a negative value indicates that amplification of that band actually impaired performance.
For frequency bands centered at 1600 Hz, 2500 Hz and 3150 Hz, the efficiency values were mostly between 1 and 0 for hearing losses (in those respective bands) up to about 60 dB. For greater losses, the efficiency values approached zero, and were sometimes negative. For the frequency band centered at 4000 Hz, the efficiency values were clustered around 0, some being slightly positive and some slightly negative. For the frequency band centered at 8000 Hz (a frequency at which the hearing losses were all above 50 dB), many of the efficiency values were negative, indicating that amplifying the very high frequencies impaired performance. Overall, these results are consistent with the idea that at least some of the subjects had high-frequency dead regions, and that, for those subjects, amplifying the high frequencies either had no effect, or made performance worse.
Turner and Cummings (1999) investigated the intelligibility of nonsense syllables in a group of ten subjects whose hearing loss at high frequencies (averaged for 4, 5 and 6.3 kHz) ranged from about 40 to 71 dB. Recognition was tested across a wide range of presentation levels. At high levels, scores reached an asymptote below 100%, i.e. performance was imperfect, but did not improve further with increasing audibility of the speech. Turner and Cummings calculated the audibility of different parts of the speech spectrum for speech at this asymptotic performance level, based on the audiogram of each subject. They found that providing audible speech at frequencies above 3 kHz tended to produce little or no improvement in recognition scores whenever the hearing loss exceeded 55 dB.
Overall, the studies described above clearly show that, when the hearing loss exceeds about 55 dB at high frequencies, amplification of high frequencies is often not beneficial. The authors of these studies have often interpreted their results in terms of damage to IHCs in the basal region of the cochlea. Specifically, it has been suggested that subjects who do not benefit from amplification have dead regions (or at least extensive damage to the IHCs) at high frequencies, while subjects who do benefit from amplification have surviving IHCs and neurons with high CFs. However, the studies all suffer from the drawback that there was no independent test for the existence of dead regions, and therefore the extent of any dead regions was unknown. I describe next a study carried out in my laboratory, with the collaboration of Deborah Vickers and Thomas Baer, in which subjects with diagnosed dead regions were tested (Moore et al., 2000a; Vickers et al., 2000; Vickers et al., 2001).
Both PTCs and the TEN test, described earlier (Moore et al., 2000b), were used to detect and define the limits of any dead regions. All subjects had high-frequency hearing loss, but some had high-frequency dead regions and some did not; generally, the subjects with dead regions had more severe high-frequency hearing losses than those without dead regions. The speech stimuli were vowel-consonant-vowel (VCV) nonsense syllables, using one of three vowels (/i/, /a/ and /u/) and 21 different consonants. In a baseline condition, subjects were tested using broadband stimuli with a nominal input level of 65 dB SPL. Prior to presentation via Sennheiser HD580 earphones, the stimuli were subjected to the frequency-gain characteristic prescribed by the “Cambridge” formula (Moore and Glasberg, 1998). This formula is intended to give speech at 65 dB SPL the same overall loudness as for a normal listener, and to make the average specific loudness (the loudness per ERB or per critical band; see Moore and Glasberg, 1997) of the speech the same for all frequencies over the range important for speech intelligibility, i.e. about 500 to 5000 Hz. Of course, this is only possible for a listener without a dead region in that frequency range. For a listener with a dead region, the specific loudness is zero for all critical bands falling within the dead region. In any case, the goal of the frequency-dependent amplification was to restore audibility as far as possible, while avoiding excessive loudness. The stimuli for all other conditions were initially subjected to this same frequency-gain characteristic. Then, the speech was lowpass filtered with various cutoff frequencies.
All subjects were given practice to familiarize them with the task. At least two lists of 63 tokens were used for each condition. These were always presented in different test sessions. The order of testing of the different cutoff frequencies was randomized in the first test session. This order was reversed for the second session to balance the effects of practice and fatigue. For subjects without dead regions, performance generally improved progressively with increasing cutoff frequency. An example is shown in the upper panel of Figure 14. This indicates that they were able to make use of high-frequency information. For subjects with dead regions, two patterns of performance were observed. For some subjects, performance initially improved with increasing cutoff frequency and then reached an asymptote (Figure 14, middle). This indicates that they were not able to make use of high-frequency information. For other subjects, performance initially improved with increasing cutoff frequency, and then worsened with further increases (Figure 14, bottom). This indicates that amplification of high frequencies impaired performance.
Percent correct scores in identifying VCV syllables for three hearing-impaired subjects, one without (top) and two with (middle, bottom) a dead region. Scores are plotted as a function of the cut-off frequency of a lowpass filter. Prior to lowpass filtering, stimuli were given the frequency-gain characteristic prescribed by the “Cambridge” formula (Moore and Glasberg, 1998). Error bars indicate ± one standard deviation across test sessions.
For the six ears without dead regions, the mean score for the broadband speech (77.2%) was significantly higher (p = 0.015) than the mean score for speech lowpass filtered at 2000 Hz (65.1%). However, for nine ears with dead regions starting below 2000 Hz, the mean scores did not differ significantly for the broadband speech (48.5%) and for the speech lowpass filtered at 2000 Hz (47.1%).
It is noteworthy that, for subjects who showed an “optimum” cutoff frequency, the best performance was achieved when that cutoff frequency was 50–100% above the estimated edge frequency of the dead region. For subjects whose performance reached an asymptote, the asymptote was reached for a cutoff frequency about 70% above the estimated edge frequency of the dead region. To assess the extent to which the subjects with dead regions were able to make use of information from frequencies just above the estimated edge frequency of the dead region, scores were compared for the lowpass filter cutoff frequency closest to the estimated edge frequency, and for the cutoff frequency closest to 1.7 times the estimated edge frequency. This analysis included all twelve ears with dead regions. The mean scores for these two conditions, 50.6 and 55.9%, respectively, differed significantly (p < 0.001). Thus, there was a significant benefit from amplifying frequencies up to 70% above the estimated edge frequencies of the dead regions.
There have been relatively few studies of the perception of non-speech sounds when the frequency components of the sounds fall within a dead region. Of those studies that do exist, the majority have used pure tone stimuli.
The Perception of Sound Quality for Pure Tones Falling in a Dead Region
Pure tones are often described as sounding highly distorted, or noise-like when they fall in a dead region (Villchur, 1973; Moore et al., 1985). This phenomenon was investigated in my laboratory by Martina Huss; only a brief report of this study has been published so far (Huss et al., 2000). Two groups of subjects with cochlear hearing loss were tested. One group of subjects was diagnosed as having dead regions using the TEN test described earlier (Moore et al., 2000b); some subjects had dead regions at low frequencies and some had them at high frequencies. In most cases, the diagnosis was confirmed by the measurement of PTCs. A second group of subjects without dead regions was also tested.
Subjects were asked to rate sounds in terms of sound quality, on a scale from 1 (clear distinct tone) to 7 (noise). Stimuli consisted of a randomly chosen sequence of pure tones of different frequencies (including frequencies falling into the dead region) and a white noise signal. The white noise signal was included as a kind of “reference”, indicating to subjects was a truly noise-like signal sounded like. The noise signal was always rated as 7. Stimuli were presented in a repeating sequence of 500-ms on (plus 10-ms onset and offset ramps) and 1000 ms off. The level for each frequency was chosen using a loudness model (Moore and Glasberg, 1997), so as to give a calculated loudness level of 20 phons. This relatively low loudness level was chosen as the sound levels required to exceed the absolute threshold were often very high for tones falling in a dead region.
Figure 15 shows typical results for subjects without (KC, top panel) and with (PJ, bottom panel) a dead region. Audiograms are indicated by the curve (referred to the left ordinate). Ratings of quality are shown as open bars (referred to the right ordinate). Except for the 125-Hz signal, low-frequency tones, falling in a region of near-normal hearing, were given low ratings, i.e. they were perceived as clear tones. In the region of increased hearing loss, ratings increased, i.e. the tones were perceived as more noise-like. However, the ratings were not markedly different for the subject with a dead region (indicated by the shaded area in the bottom panel) and the subject without a dead region.
Results of the clarity-rating task for two subjects, one without (KC, top panel) and one with (PJ, bottom panel) a high-frequency dead region (indicated by the shaded area). Audiograms are indicated by the curves (referred to the left ordinate). Ratings of quality are shown as open bars (referred to the right ordinate). Error bars indicate ± 1 SD.
Figure 16 shows results from a subject (AW) with extensive dead regions in the right ear and a moderate high-frequency loss without any dead region in the left ear. For the left ear (top panel), ratings increased at high frequencies where the hearing loss was greater. For the right ear (bottom panel), ratings for tones falling within the dead regions varied markedly across frequency.
Results of the clarity-rating task for a subject (AW) with extensive dead regions in the right ear (bottom; dead regions indicated by shaded areas) and a moderate high-frequency loss without any dead region in the left ear (top). Otherwise as Figure 15.
Considering the results for all the subjects tested (not shown here), there was a trend for noise-like percepts evoked by pure tones to be associated with dead regions. However, the data shown in Figures 15 and 16 indicate that the association is not consistent. Reports that pure tones sound noise-like may be taken as a hint that a dead region is present, but ratings of the clarity of the tonal percept cannot be used as a reliable indicator of dead regions.
Several studies of the perceived pitch of low-frequency sinusoids have been conducted for people with low-frequency dead regions. For such cases, a low-frequency pure tone cannot produce maximum neural excitation at the CF corresponding to its frequency. The peak in the neural excitation pattern must occur at CFs higher than the frequency of the signal. If the place theory of pitch is correct, this should lead to marked upward shifts in the pitch of the tone. In fact, this does not always happen, although it does happen sometimes.
Florentine and Houtsma (1983) obtained pitch matches between the two ears of a subject with a low-frequency dead region in one ear only. They presented the stimuli at levels just above absolute threshold, to minimize the spread of excitation along the basilar membrane. Pitch shifts between the two ears were small. Turner et al. (1983) studied six subjects with low-frequency cochlear losses. Three of their subjects gave PTCs with tips close to the signal frequency; they presumably had functioning IHCs with CFs close to the signal frequency. The other three subjects gave PTCs with tips well above the signal frequency; they presumably had low-frequency dead regions. Pitch perception was studied either by pitch matching between the two ears (for subjects with unilateral losses) or by octave matching (for subjects with bilateral losses, but with some musical ability). The subjects with dead regions gave results similar to those of the subjects without dead regions; no distinct pitch anomalies were observed.
More recently, a similar study was conducted in my laboratory by Martina Huss; only a brief report of this study has been published so far (Huss et al., 2000). Subjects were the same as for the task involving ratings of clarity, described above. Two tasks were used, a pitch-matching task and an octave-matching task. For the pitch-matching task, subjects were asked to match the perceived pitch of a pure tone with that of another fixed-frequency pure tone. The two tones were presented alternately. Matches were made across ears, to obtain a measure of diplacusis, and within one ear, to estimate the reliability of matching. For the octave-matching task, subjects were asked to adjust a tone of variable frequency so that it sounded one octave higher or lower than a fixed reference tone. Only a few subjects were able to perform this task reliably. The level for each frequency was chosen using a loudness model (Moore and Glasberg, 1997), so as to give a calculated loudness level of either 50 or 60 phons.
Results of the pitch-matching task for subject AW (the subject with extensive dead regions in the right ear and a moderate high-frequency loss without dead region in the left ear) are shown in Figure 17. Each x denotes one match, and means are shown by open circles. Matches within his better ear (top) were reasonably accurate at low frequencies, but became less accurate at high frequencies. Matches within his worse ear (middle), which has extensive dead regions, were more erratic, indicating a less clear pitch percept. Matches across ears, with the fixed tone in his worse ear (bottom), showed considerable variability, but also some consistent deviations. A fixed tone of 0.5 kHz in the worse ear was matched with a tone of about 3.5 kHz in the better ear. Generally, the matched frequency lay above the fixed frequency, for all fixed frequencies up to about 4 kHz, indicating upward pitch shifts in the worse ear.
Results of the pitch-matching task for subject AW, who had extensive dead regions in his worse ear, shown by the shaded areas, and a moderate high-frequency loss without any dead region in his better ear. Each x denotes one match, and means are shown by open circles. Matches were made within his better ear (top), within his worse ear (middle), and across ears (bottom).
Results of the pitch-matching and octave-matching tasks for a subject with an extensive high-frequency dead region are shown in Figure 18 (results for one ear only; the other ear was “dead”). The dead region was estimated to start at about 1.2 kHz. Pitch matches (top) were reasonably accurate for frequencies up to 1.25 kHz, and then became much more erratic, indicating that a clear pitch percept was not obtained at frequencies above 1.25 kHz. Octave matches with the lower tone fixed in frequency (middle) resulted in frequency ratios around 2 (the “expected” value) for fixed frequencies up to 0.5 kHz. For a fixed frequency of 1 kHz, the upper tone was adjusted to about 1.4 kHz; when the upper tone fell within the dead region its pitch was higher than “normal”. Octave matches with the upper tone fixed in frequency (bottom) resulted in frequency ratios around (but a little above) 0.5 (the “expected” value) for fixed frequencies up to 1 kHz. For fixed frequencies of 1.76 and 2 kHz, octave matches clearly deviated from a ratio of 0.5. For tones whose frequencies fell well within the dead region, the perceived pitch was shifted upwards, although it was also unclear.
Results of the pitch-matching and octave-matching tasks within one ear for subject RC, who had a high-frequency dead region starting at about 1.2 kHz (the other ear was “dead”). Matches are shown in the top panel. Octave matches were made with the lower tone fixed in frequency (middle) and with the upper tone fixed in frequency (bottom).
Measures of frequency discrimination also suggest that tones with frequencies falling in a dead region do not evoke a clear pitch. McDermott et al. (1998) measured frequency difference limens (DLs) as a function of center frequency for subjects with near-normal hearing at low frequencies with a steep transition to severe or profound loss at higher frequencies. It is likely that all of these subjects had dead regions at higher frequencies. For tones falling in the frequency range of near-normal hearing, the DLs were typically about 2–3% of the center frequency. For tones falling well within the dead region, the DLs increased to about 10%. Interestingly, the DLs sometimes decreased slightly for frequencies just below the presumed boundary of the dead region, which McDermott et al. suggested might be the result of cortical over-representation of CFs just below the dead region (Irvine and Rajan, 1995).
Taken together, the results of studies of pitch perception using people with dead regions indicate the following:
Pitch matches (of a tone with itself, within one ear) are often erratic, and frequency discrimination is poor, for tones with frequencies falling in a dead region. This indicates that such tones do not evoke a clear pitch sensation.
Pitch matches across the ears of subjects with asymmetric hearing loss, and octave matches within ears, indicate that tones falling within a dead region sometimes are perceived with a near-“normal” pitch and sometimes are perceived with a pitch distinctly different from “normal”.
The shifted pitches found for some subjects indicate that the pitch of low-frequency tones is not represented solely by a temporal code. Possibly, there needs to be a correspondence between place and temporal information for a “normal” pitch to be perceived (Evans, 1978; Loeb et al., 1983; Srulovicz and Goldstein, 1983).
McDermott et al. (1998) studied five subjects with near-normal hearing at low frequencies with a steep transition to severe or profound loss at higher frequencies. It is likely that all of these subjects had dead regions at higher frequencies. McDermott et al. obtained equal-loudness contours using a reference tone presented at 500 Hz, which was within the region of near-normal hearing for all subjects. For frequencies close to the upper limit of hearing (i.e. well within the dead region), the equal-loudness contours were “squeezed” together, indicating a very rapid growth of loudness with increasing sound level. In one case (their subject S4), when the reference 500-Hz tone was increased in level by 60 dB, a 1000-Hz matching tone had to be increased in level by only about 10 dB to produce a corresponding increase in loudness. This represents a kind of “super-recruitment”, a very rapid growth of loudness with increasing level, once the detection threshold is exceeded. We have attempted to account for these results using the loudness model of Moore and Glasberg (1997), but were unsuccessful, even though that model can take dead regions into account in generating predictions of loudness.
The discrepancy between the data and the model might be accounted for in several ways. Firstly, the model might be wrong in its basic assumptions or implementation. Secondly, the super-recruitment may arise because of cortical re-organization, leading to an over-representation of CFs just below the boundary of the dead region (Irvine and Rajan, 1995). Thirdly, the loudness judgments may have been influenced by the subjective quality of the tone falling in the dead region. The perceived amount of distortion may increase rapidly with increasing level. Subjects may have associated the percept of distortion with a high loudness. Whatever the reason for the super-recruitment, it is clear that the usable dynamic range is often very narrow for tones falling in a dead region.
Implications of Dead Regions for Fitting Hearing Aids
Choosing the Appropriate Frequency-Gain Characteristic for “Conventional” Hearing Aids
The studies on speech perception reviewed above suggest that, for people with high-frequency dead regions:
There is little or no benefit to speech discrimination from amplifying frequencies well inside a dead region.
There may be some benefit in amplifying frequencies up to 50–100% above the estimated edge frequency of the dead region. However, this conclusion is based mainly on our study of the intelligibility of speech in quiet. The “optimum” cutoff frequency may be different for speech presented in background noise. Also, it is not clear what the optimum cutoff frequency may be for listening to music, or for detecting and identifying environmental sounds.
It should be emphasized that, for patients without high-frequency dead regions, amplification of the high frequencies can be (and usually is) beneficial (Skinner and Miller, 1983; Vickers et al., 2000). Hence, before deciding what form of amplification should be provided for a patient with high-frequency hearing loss, it is important to determine whether that patient has a high-frequency dead region, and, if so, what its extent is. I recommend the TEN test for this purpose.
For a patient with a dead region at high frequencies, there may be several benefits of reducing the gain at high frequencies. Firstly, this may actually lead to improved speech intelligibility, as described above. Secondly, it may reduce problems associated with acoustic feedback. Thirdly, it may reduce distortion in the hearing aid, especially intermodulation distortion. Finally, it allows the dispenser to concentrate efforts on providing appropriate amplification over the frequency range where there is useful residual hearing.
The situation for patients with dead regions at low frequencies is less clear cut. As described above, there is evidence that patients with low-frequency dead regions do not extract useful information about speech from IHCs/neurons tuned to low frequencies. However, such patients do appear to have some ability to extract information from low-frequency components of speech via “channels” tuned to higher frequencies, if those components are suitably amplified. If the strategy is adopted of not amplifying low frequencies, as advocated by Halpin et al. (1994), the question still remains as to the frequency range over which amplification should be applied (assuming that hearing is impaired at medium and high frequencies). Possibly, amplification should be applied over a frequency range extending somewhat into the dead region. Further research in this area is clearly needed.
If a patient has a hearing loss over the frequency range where there are surviving IHCs and neurons, consideration should be given to providing a hearing aid with some form of compression or automatic gain control (AGC). There is now considerable evidence that AGC is beneficial in allowing patients to deal with the wide range of sound levels that they encounter in everyday life (Laurence et al., 1983; Moore et al., 1992; Moore, 1993; Hickson, 1994; Dillon, 1996; Moore, 1998). Methods for the initial fitting of multi-channel compression hearing aids have been developed by several research groups. Some of these methods, for example the “Camfit” methods developed in my laboratory (Moore et al., 1999a; Moore, 2000), are based on a loudness model (Moore and Glasberg, 1997), and have the default assumption that no dead region exists. Before using such methods, a test for the diagnosis of dead regions should be applied (such as the TEN test) whenever the audiogram indicates the possibility of a dead region. When a dead region is diagnosed, the gains recommended by the fitting method should be applied only for frequencies where there is not a dead region, and perhaps for frequencies extending 50–100% inside any dead region.
Another method for the fitting of hearing aids, the NAL-NL1 method (Dillon, 1999), is partly based on the same loudness model, but is also based on empirical data on the effects of amplification of speech for hearing-impaired listeners with varying degrees of hearing loss (Ching et al., 1998). These data show that, on average, people with high-frequency hearing loss get progressively less benefit from amplification of the high frequencies as the hearing loss increases; some of these data were reviewed above. This is taken into account in the fitting method, by reducing the high-frequency gain progressively below the value needed to restore audibility as the hearing loss increases.
In my view, this approach, based on data averaged across listeners, is probably inappropriate, since some of the listeners had dead regions and some did not. For listeners without dead regions, amplification of the high frequencies so as to restore audibility usually leads to improved intelligibility (Skinner and Miller, 1983; Vickers et al., 2000). For listeners with dead regions at high frequencies, amplification of high frequencies so as to restore audibility will not be beneficial. Gains based on averaging data across these two classes of listeners are likely to be inappropriate for both classes.
Transposition and Frequency Compression in Hearing Aids
Many researchers over the years have suggested the possibility of frequency transposition or frequency compression as a method of making more speech information available to patients with dead regions. Most commonly, downward frequency transposition or compression has been tried for patients with high-frequency dead regions (Johansson, 1966; Velmans and Marcuson, 1983; Posen et al., 1993; Parent et al., 1997; McDermott et al., 1999), although upward transposition has also been used (Berlin, 1982).
There are several potential problems with transposition and compression. Firstly, there is likely to be a limit to the amount of information that can be “squeezed” into the limited region of residual hearing; there is a danger of “overloading” that region. Secondly, the transposed or compressed information is presented to the “wrong” place in the cochlea, and, as noted earlier, this may lead to difficulties in interpreting that information (Shannon et al., 1998). Certainly, an extended learning period may be required to make effective use of information from the transposed/compressed frequencies (Rosen et al., 1999). Thirdly, when background noise is present, portions of that noise that were previously inaudible may be transposed to a frequency region where they are more audible, and this may offset any advantage that would otherwise be gained from the transposition.
So far, hearing aids incorporating frequency transposition and/or compression have not found widespread acceptance. However, promising results have been found in some studies of commercially available devices (Foust and Gengel, 1973; Parent et al., 1997). The limited benefit demonstrated so far may partly have occurred because the transposition/compression aids have been fitted to patients without clear knowledge of the extent of the dead regions. Hopefully, the availability of the TEN test will lead to more accurate diagnosis of dead regions, and hence to the possibility of better fitting. Clearly, more research in this area is needed.
Concluding Remarks
I summarize here some of the main conclusions that can be drawn from this review:
Dead regions may be relatively common in people with moderate-to-severe sensorineural hearing loss.
Dead regions cannot be reliably diagnosed from the audiogram. However, dead regions may be commonly associated with: (a) A hearing loss more than 90 dB at high frequencies or 75–80 dB at low frequencies; (b) A hearing loss of 40–50 dB at low frequencies with near-normal hearing at medium and high frequencies (perhaps indicating a low-frequency dead region); (c) A hearing loss greater than 50 dB at low frequencies with somewhat less hearing loss at higher frequencies (also perhaps indicating a low-frequency dead region); (d) A hearing loss increasing rapidly (more than 50 dB/octave) with increasing frequency (perhaps indicating a high-frequency dead region). Whenever one of these patterns is found, further tests, such as the TEN test, should be performed to establish whether or not a dead region is present.
Psychophysical tuning curves (PTCs) provide a useful way of detecting dead regions and defining their boundaries. When the tip of the PTC is shifted markedly from the signal frequency, then, to a first approximation, the frequency at the tip indicates the boundary of the dead region. However, the frequency at the tip of the PTC may not give an accurate estimate of the boundary when the shift is small (less than a few hundred Hertz). The determination of PTCs is probably too time-consuming to be used for routine diagnosis of dead regions in clinical practice.
The measurement of detection thresholds for pure tones in threshold equalizing noise (TEN) provides a simple method for clinical diagnosis of dead regions. A dead region at a specific frequency is indicated by a threshold that is 10 dB or more above absolute threshold and 10 dB or more above the noise level per ERB. However, a failure to meet one or both of these criteria does not necessarily imply absence of a dead region.
Pure tones with frequencies falling in a dead region do not evoke clear pitch sensations (pitch matching is highly variable) and the perceived pitch is sometimes, but not always, different from “normal”. However, ratings of pitch clarity cannot be used as a reliable indicator of a dead region.
Amplification of frequencies well inside a high-frequency dead region usually does not improve speech intelligibility, and may sometimes impair it. However, there may be some benefit for speech intelligibility in amplifying frequencies up to 50–100% above the estimated edge frequency of a high-frequency dead region. The optimum upper limit of amplification for sounds other than speech remains to be determined.
The optimal form of amplification for people with low-frequency dead regions remains somewhat unclear. There may be some benefit from avoiding the amplification of frequencies well inside a dead region.
Patients with extensive dead regions are likely to get less benefit from hearing aids than patients without dead regions. Even with a hearing aid, patients with extensive dead regions are likely to have relatively poor speech discrimination.
For patients with diagnosed dead regions at high frequencies, consideration should be given to use of a hearing aid incorporating frequency transposition and/or compression.
Obtaining the Ten Test CD
To obtain the test CD, people in the Americas and Asia should contact:
Starkey Laboratories, Inc.
Department of Advanced Research
6700 Washington Ave. So.
Eden Prairie, MN USA 55344
Attention: Paula Smith
Telephone (within the USA) 1 800 328 8602
People in Europe should contact:
Starkey Laboratories, Inc.
William F. Austin House
Bramhall Technology Park
Pepper Road, Hazel Grove
Stockport SK7 5BX, United Kingdom
Telephone:
(within mainland Europe) + 44 161 483 2200
(within the UK) 0161 483 2200
Fax:
(within mainland Europe) + 44 161 483 9833
(within the UK) 0161 483 9833
Footnotes
Notes
The work described in this paper was supported mainly by the MRC (UK), with additional support from Starkey (USA), Defeating Deafness (UK) and the RNID (UK)
Acknowledgments
I thank Martina Huss, Deborah Vickers, Thomas Baer, Brian Glasberg and Joseph Alcántara for their valuable contributions to the work described in this paper. I also thank Hedwig Gockel, Martina Huss, Todd Ricketts, Michael Stone and two anonymous reviewers for helpful comments on an earlier version of this paper.