Challenges and Recent Developments in Hearing Aids: Part I. Speech Understanding in Noise,Microphone Technologies and Noise Reduction Algorithms

3.1.1.1. How They Work

First-order directional microphones are implemented with either a single-microphone design or a dual/twin-microphone design. In the single-microphone design, the directional microphone has an anterior and a posterior microphone port. The acoustic signal entering the posterior port is acoustically delayed and subtracted from the signal entering the anterior port at the diaphragm of the microphone.

The rationale is that if a sound comes from the front, it reaches the anterior port first and then reaches the posterior port a few milliseconds later. Since the sound in the posterior port is delayed by the traveling time between the two microphone ports (i.e., external delay) and the acoustic delay network (i.e., the internal delay), the sound from the front is minimally affected. Therefore, the directional microphones have high sensitivity to sounds from the front. However, if a sound comes from the back, it reaches the posterior port first and continues to travel to the anterior port. If the internal delay equals the external delay, sounds entering the posterior port and the anterior port reach the diaphragm at the same time but on opposite sides of the diaphragm, thus they are cancelled. Therefore, the sensitivity of the directional microphone to sounds from the back is greatly reduced.

The sensitivity of the directional microphone to sounds coming from different azimuths is usually displayed in a polar pattern. Directional microphones exhibit four distinct types of polar patterns: bipolar (or bidirectional, dipole), hypercardioid, supercardioid, and cardioid (Figure 1A). The least sensitive microphone locations (i.e., nulls) of these polar patterns are at different azimuths relative to the most sensitive location (0° azimuth). Notice that these measurements are made when the directional microphones are free-hanging in a free field where the sound field is uniform, free from boundaries, free from the disturbance of other sound sources, and nonreflective. When the directional microphones are measured in three dimensions, their sensitivity patterns to sounds from different locations are called directional sensitivity patterns (Figure 1B).

OPEN IN VIEWER

Figure 1.

(A) The relationship between the ratio of the internal and external delays and the polar patterns. (Reprinted with permission from Powers and Hamacher, Hear J, 55[10], 2002). (B) The directional sensitivity patterns of directional microphones exhibiting cardioid, hypercardioid, and supercardioid patterns. A commercial stand directional microphone is placed at the center of the directional sensitivity pattern.

The directional sensitivity patterns of directional microphones are generated by altering the ratio between the internal and external delays. The internal delay is determined by the acoustic delay network placed close to the entrance of the back microphone port. The external delay is determined by the port spacing between the front and back ports, which in turn, is determined by the available space and considerations of the amount of low-frequency gain reduction and the amount of high-frequency directivity (Figure 2).

OPEN IN VIEWER

Figure 2.

(A) The relationship between the port spacing (p) and the amount of low-frequency cut-off for directional microphones that use the delay-and-subtract processing. As port spacing decreases, the cut-off frequency for low-frequency gain reduction increases. This is because sound pressures picked up by the two microphone ports/two omni-directional microphones are subtracted at two adjacent points. As frequency decreases, the wavelength increases, the differences between the two points decreases, and the resultant microphone output becomes smaller after the subtraction. Thus, the cut-off frequency for low-frequency roll-off increases as the microphone port spacing decreases. (B) The relationship between the port spacing (p) and the amount of high-frequency directivity index (DI). As port spacing decreases, the high-frequency directivity index increases. This occurs because as the wavelength of the incoming signal approaches the port spacing, directionality breaks down. The smaller the port spacing, the higher the frequency at which the directionality breaks down (AI-DI = articulation index weighted directivity index). (Courtesy of Oticon, reprinted with permission).

In the dual-microphone design, the directional microphones are composed of two omni-directional microphones that are matched in frequency response and phase (Figure 3). The two omnidirectional microphones are combined by using delay-and-subtract processing, similar to single-microphone directional microphones. The electrical signal generated from the posterior microphone is electrically delayed and subtracted from that of the anterior microphone in an integrated circuit (Buerkli-Halevy, 1987; Preves, 1999; Ricketts and Mueller, 1999). By varying the ratio between the internal and external delays, the polar patterns of the dual-microphone directional microphones can also generate bipolar, cardioid, hypercardioid, or supercardioid patterns.

OPEN IN VIEWER

Figure 3.

A figurative example of how the first-order dual-microphone directional microphone is implemented. The polar pattern of the directional microphone depends on the ratio of the internal and external delays. (Reprinted (modified) with permission from Thompson SC, Hear J 56[11], 2003).

Although the performances of single-microphone and dual-microphone directional microphones are comparable, most of the high-performance digital hearing aids use dual-microphone directional microphones because of their flexibility. Single-microphone directional microphones have fixed polar patterns after being manufactured because neither the external delay nor the internal delay can be altered. However, dual-microphone directional microphones can have variable polar patterns because their internal delays can be varied by signal processing algorithms. The ability to vary the polar pattern after the hearing aid is made opens doors to the implementation of advanced signal processing algorithms (e.g., adaptive directional microphone algorithms).

The directional effect of the directional microphones can be quantified in several ways:

For a review on the design and evaluation of first-order directional microphones, please refer to the reviews by Ricketts (2001) and Valente (1999).

3.1.1.2. Updates on the Clinical Verification of First Order Directional Microphones

Many factors affect the benefits of directional microphones. Research studies on the effect of directional microphones on speech recognition conducted in laboratory settings showed a large range of SNR-50 improvement, from 1 to 16 dB. The amount of benefit experienced by the hearing aid users depends on the directivity index of the directional microphone; the number, the placement, and the type of noise sources; the room and environmental acoustics, relative distance between the talker and listener, location of the noise relative to the listener, and vent size, among others (Beck, 1983; Hawkins and Yacullo, 1984; Gravel et al., 1999; Kuk et al., 1999; Nielsen and Ludvigsen, 1978; Preves et al., 1999; Ricketts, 2000a; Ricketts and Dhar, 1999; Studebaker et al., 1980; Valente et al., 1995; Wouters et al., 1999).

Normally, the higher the directivity index and AI-DI, the higher the directional benefit provided by the directional microphones (Ricketts, 2000b, Ricketts et al., 2001). Studies by various researchers have also shown that the directivity index or AI-DI can be used to predict the improvements in SNRs-50 provided by the directional microphones with multiple noise sources or relatively diffuse environments (Laugesen and Schmidtke, 2004; Mueller and Ricketts, 2000; Ricketts, 2000a).

The amount of directional benefit of the hearing aids is affected by the number of noise sources and the testing environment. Studies conducted with one noise source placed at the null of the directional microphones (Agnew and Block, 1997; Gravel et al., 1999; Lurquin and Rafhay, 1996; Mueller and John, 1979; Valente et al., 1995; Wouters et al., 2002) showed greater SNR-50 improvements than studies conducted with multiple noise sources or with testing materials recorded in real-world environments (Killion et al., 1998; Preves et al., 1999; Pumfort et al., 2000; Ricketts, 2000a; Ricketts and Dhar, 1999; Valente et al., 2000). In general, 3 to 5 dB of improvement in the SNR-50 is reported in real-world environments with multiple noise sources (Amlani, 2001; Ricketts et al., 2001; Valente et al., 2000; Wouters et al., 1999).

In addition, greater improvements are generally observed in less reverberant environments than in more reverberant environments (Hawkins and Yacullo, 1984; Killion et al., 1998; Ricketts, 2000b; Ricketts and Dhar, 1999; Studebaker et al., 1980; Ricketts and Henry, 2002). Reverberation reduces directional effects because sounds are reflected from surfaces in all directions. The reflected sounds make it impossible for directional microphones to take advantage of the spatial separation between speech and noise. Research studies have also shown that directional microphones are more effective if speech, noise, or both speech and noise are within the critical distance (Ricketts, 2000a; Leeuw and Dreschler, 1991; Ricketts and Hornsby, 2003). Critical distance is the distance at which the level of the direct sound is equal to the level of the reverberant sound. Within the critical distance, the level of the direct sound is higher than the level of the reverberant sound.

Further, the directional effect at the low-frequency region reduces as the vent size increases because vents tend to reduce the gain of the hearing aid below 1000 Hz and allow unprocessed signals from all directions to reach the ear canal. However, when the weightings of articulation index are considered, the decrease in AI-DI was only about 0.4 dB for each 1-mm increase in vent size up to a diameter of 2 mm (Ricketts, 2001; Ricketts and Dittberner, 2002). Although a larger decrease of AI-DI (i.e., 0.8 dB) was observed when the vent size increased from 2 mm to open fitting, the open earmold fitting would still have about a 4 dB higher AI-DI than the omni-directional mode. In general, vents have the greatest effect on hearing aids with high directivity indexes at low frequencies (Ricketts, 2001).

Factors that do not affect the benefit of directional microphones are compression and hearing aid style (Pumfort et al., 2000; Ricketts, 2000b; Ricketts et al., 2001). At the first glance, the actions of compression and directional microphones seem to act in opposite directions, i.e., directional microphones reduce background noise which is usually softer than speech while compression amplifies softer sounds more than louder sounds. In practice, however, sounds from multiple sources occur at the same time and the gain of the compression circuit is determined by the most dominant source or the overall level. If both speech and noise occur at the same instance with a positive SNR, the gain of the hearing aid is determined by the level of the speech, not the noise. Research studies comparing the directional benefits of linear and compression hearing aids did not show any difference in speech understanding ability of hearing aid users if speech and noise coexist at the same instance (Ricketts et al., 2001).

Another factor that does not affect the performance of directional microphones is the hearing aid style (Pumfort et al., 2000; Ricketts et al., 2001). Previous research studies have shown that the omni-directional microphones of the in-the-ear hearing aids have higher directivity indexes than do behind-the-ear hearing aids because of the pinna effect (Fortune, 1997, Olsen and Hagerman, 2002) and the SNRs-50 of subjects also concur with this finding (Pumfort et al., 2000). However, SNRs-50 of subjects were not significantly different for the directional microphones of the two hearing aid styles (Pumfort et al., 2000; Ricketts et al., 2001). This indicated that directional microphones provide less improvement in speech understanding in an in-the-ear hearing aid than a behind-the-ear hearing aid. In other words, although the omni-directional microphones of in-the-ear hearing aids are more directional than the omni-directional microphones of the behind-the-ear hearing aids, the performance of the directional microphones implemented in both hearing aid styles was not significantly different (Ricketts, 2001).

Most laboratory tests have shown measurable directional benefits and many hearing aids users in field evaluation studies also report perceived directional benefit. However, a number of recent field studies reported that a significant percentage of hearing aid users might not perceive the benefits of directional amplification in their daily lives even if the signal processing, venting, and hearing aid style are kept the same in the field trials and laboratory tests (Cord et al., 2002; Mueller et al., 1983; Ricketts et al., 2003; Surr et al., 2002; Walden et al., 2000).

According to the researchers, the possible reasons for the discrepancies can be attributed to the relative locations of the signal and noise, acoustic environments, the type and location of noise encountered, subjects’ willingness to switch between directional and omni-directional microphones, and the percentage of time the use of directional microphone is indicated, among others. (Cord et al., 2002, Surr et al., 2002, Walden et al., 2000; Walden et al., 2004).

Specifically, directional microphones are designed to be more sensitive to sounds coming from the front than sounds coming from other directions. Many laboratory tests showing the benefit of directional microphones were conducted with speech presented at 0° azimuth and noise from the sides or the back with both speech and noise in close proximity to the hearing aid user. However, hearing aid users reported that the desired signal did not come from the front in as much as 20% of the time in daily life (Walden et al., 2003). Studies have indicated that when speech comes from directions other than the front, the use of directional microphone may have a positive, neutral, or negative effect on speech intelligibility, especially for low-level speech from the back (Kuk, 1996; Kuk et al., 2005, Lee et al., 1998; Ricketts et al., 2003).

Two other possible reasons for the discrepancies between laboratory tests and field trials are the acoustic environments, and the type(s) and location(s) of noise encountered. Most laboratory tests are conducted in environments with a reverberation time of less than 600 milliseconds (Amlani, 2001). A wide range of reverberation environments that often have higher reverberation times may be encountered in daily life, however. As directional benefits diminish with increase in reverberation, hearing aid users may not be able to detect the benefits of directional microphones in their everyday life. In addition, the use of non-real-world noise in the laboratory (e.g., speech spectrum noise) at fixed, examiner-determined locations may have exaggerated the benefits of directional microphones.

The need for the user to switch between omni-directional microphones and directional microphones and the percentage of time/situations that the use of directional microphone is indicated in daily life can also partly account for the laboratory and field evaluation differences. Cord and colleagues (2002) reported that about 23% of the subjects stated that they left their hearing aids in the default omni-directional mode during the field trials because they did not notice much difference in the first few trials of directional microphones. Further, Cord and colleagues reported that subjects who actively engaged in switching between the omni-directional and directional microphones reported that they only used directional microphones about 22% of the time. This indicated that omni-directional microphones were sufficient in 78% of daily life, and subjects may not have adequate directional microphones usage time to realize the benefit of directional microphones.

Cord and colleagues (2004) have also investigated the predictive factors differentiating the subjects who regularly switched between the omni-directional and directional modes and those who left the hearing aids in the default position in a subsequent study. They reported that the two groups did not significantly differ in their degree or configuration of hearing loss, hearing aid settings, directional benefits that they receive when tested in the test booth, or the likelihood to encounter situations where bothersome background noise occurs. In other words, there is no ensured evidence that can be used to predict which hearing aid users will switch between the omni-directional and directional microphones versus those who will leave the hearing aids in the default omni-directional mode. In addition, previous studies also failed to predict directional benefits based on hearing aid users’ audiometric testing results (Jespersen and Olsen, 2003; Ricketts and Mueller, 2000).

3.1.1.3. Updates on the Limitations of First-Order Directional Microphones

With the increase in directional microphone usage in recent years, the limitations of directional microphones have become more apparent to the hearing aid engineers and audiology community. These limitations of directional microphones include relatively higher internal noise, low-frequency gain reduction (roll-off), higher sensitivity to wind noise, and reduced probability to hear soft sounds from the back (Kuk et al., 2005, Lee et al., 1998; Ricketts and Henry, 2002; Thompson, 1999).

Two factors contribute to the problem of higher internal noise for the dual-microphone directional microphones. First, the internal noise of the modern omni-directional microphones is about 28 dB SPL. When two omni-directional microphones are combined to make a dual-microphone directional microphone in the delay-and-subtract process, the internal noise of dual microphones is about 3 dB higher than the internal noise of omni-directional microphones (Thompson, 1999). This internal noise is normally masked by environmental sounds and is inaudible to hearing aid users, even in quiet environments. However, the problem arises when a hearing aid manufacturer tries to accommodate the second factor, low-frequency roll-off.

The low-frequency roll-off occurs when low-frequency sounds reaching the two omni-directional microphones are subtracted at similar phase. The amount of low-frequency roll-off is about 6 dB/octave for first-order directional microphones (Thompson, 1999; Ricketts, 2001). The perceptual consequence of the low-frequency roll-off is “tinny” sound quality and under-amplification of low-frequency sounds for hearing aid users with low-frequency hearing loss (Ricketts and Henry, 2002; Walden et al., 2004).

The common practice to solve this problem is to provide low-frequency equalization so that the low-frequency responses of the directional microphones are similar to that of the omni-directional microphones. Unfortunately, by matching the gain between omni-directional and directional modes, the internal microphone noise is also amplified. Some hearing aid users may find this increase in microphone noise objectionable, especially in quiet environments (Lee and Geddes, 1998; Macrae and Dillon, 1996).

Two practices are adopted to circumvent this dilemma. First, instead of fully compensating for the 6-dB/octave low-frequency roll-off, hearing aid manufacturers may decide to provide partial low-frequency compensation (i.e., 3 dB/octave). Second, the consensus in the audiology community is to stay in the omni-directional mode in quiet environments. Field studies have also shown that subjects either preferred the omni-directional mode or showed no preference between the two modes in quiet environments (Mueller et al., 1983; Preves et al., 1999; Walden et al., 2004).

Directional microphones are also more susceptible to wind noise because they have a higher sensitivity to near field signals. When the wind curves around the head, turbulence is created very close to the head. As directional microphones have higher sensitivity to sounds in near field (i.e., sounds within 30 cm), the wind noise level picked up by the directional microphones can be as much as 20 to 30 dB higher than that picked up by an omni-directional microphone (Figure 4) (Kuk et al., 2000; Thompson, 1999). Because wind noise has dominant energy at low frequencies, the negative effect of wind noise is further exacerbated if the directional microphone has low-frequency gain compensation. Again, the strategy is to use omni-directional microphone mode should wind noise be the dominant signal at the microphone input. In addition, some algorithms automatically reduce low frequency amplification when wind noise is detected (Siemens Audiology Group, 2004).

OPEN IN VIEWER

Figure 4.

Directional microphones (solid line) have higher outputs for wind noise than omni-directional microphones (dotted line). (Original data from Dillon et al., 1999. Reprinted with permission from Kuk et al., Hear Rev 7[9], 2000).

Although a design objective, it can be said that a limitation of directional microphones is that they are less sensitive to speech or environmental sounds coming from the back hemisphere, especially at low levels (Kuk et al., 2005; Lee et al., 1998). Directional microphones should be used with caution in environments in which audibility to sounds or warning signals from the back hemisphere is desirable.

3.1.1.4. Working with Directional Microphones

Despite these limitations, directional microphones are currently the most effective noise reduction strategy (second to personal FM or infrared systems) available in hearing aids. Several cautions should be exercised when clinicians fit directional microphones:

First, the performance of directional microphones decreases near reflective surfaces, such as a wall or a hand, or in reverberant environments. Hearing aid users therefore need to be counseled to move away from reflective surfaces or to converse at a place with less reverberation, if possible.

Second, the polar patterns of a directional hearing aid and the locations of the nulls when the hearing aid is worn on the user's head can be very different from an anechoic chamber measurement in which the directional microphone is free-hanging in space (Chung and Neuman, 2003; Neuman et al., 2002). Depending on the hearing aid style, the most sensitive angle of the first-order directional microphones may vary from 30° to 45° for in-the-ear hearing aids to 90° for behind-the-ear hearing aids (Foutune, 1997, Neuman et al, 2002; Ricketts, 2000b). If possible, clinicians need to use the polar patterns measured when the hearing aids are worn in the ear so they can counsel the hearing aid users to position themselves so that the most sensitive direction of their directional microphone points to the direction of the desired signal and the most intense noise is located at the direction with the least sensitivity, if possible.

Third, clinicians need to be aware that some hearing aids automatically provide low-frequency compensation for the directional microphone mode. Others require the clinician to select the low-frequency compensation in the fitting software. Clinicians also need to determine if low-frequency compensation for the directional microphone mode is appropriate given the hearing aid user's listening needs.

Fourth, Walden and colleagues (2004) recently reported that hearing aid users who actively switch between the omni-directional and directional microphones preferred the omni-directional mode more in relatively quiet listening situations. When noise existed, they preferred the omni-directional mode when the signal source was relatively far away. On the other hand, hearing aid users tended to prefer the directional mode in noisy environments, when speech comes from the front, or when the signal is relatively close to them. Walden and colleagues also noted that counseling hearing aid users to switch to the appropriate microphone mode might increase the success rate of directional hearing aid fitting.

Fifth, although a number of studies have shown that children can also benefit from directional microphones to understand speech in noise (Condie et al., 2002; Gravel e al., 1999; Bohnert and Brantzen, 2004), the use of directional microphones that require manual switching in very young children should be cautioned. Very young children who are starting to learn the auditory, speech, and language skills need every opportunity to access auditory stimuli. As directional microphones attenuate sounds from the sides and back, they may reduce the incidental learning opportunities that may help children acquire or develop speech and language skills. In addition, young children probably will not be able to switch between microphone modes requiring parents or caregivers to effectively assume this responsibility among other care-giving liabilities.

As mentioned before, always listening in the directional mode may reduce the chance of detecting warning signals or soft speech from behind, which is crucial to a child's safety. The American Academy of Audiology recommended the use of directional microphones on children with caution, especially on young children who cannot switch between the directional and omni-directional modes (American Academy of Audiology, 2003).

3.1.2.1. How They Work

Most of the adaptive directional microphones implemented in commercially available hearing aids are first-order directional microphones. The physical construction of adaptive directional microphones is identical to that of the dual-microphone directional microphones. The difference is that the signal processing algorithm of the adaptive directional microphones can take advantage of the independent microphone outputs of the omni-directional microphones and vary the internal delay of the posterior microphone. As mentioned before, the polar pattern of a directional microphone can be changed by varying the ratio of the internal and external delays. Because the external delay (determined by the microphone spacing) is fixed after the hearing aid is manufactured, the ratio of the internal and external delays can be changed by varying the internal delay of the posterior microphone. When the ratio is changed from 0 to 1, the polar pattern is varied from bidirectional to cardioid (Powers and Hamacher, 2004).

Ideally, adaptive directional microphones should always adopt the polar pattern that places the nulls at the azimuths of the dominant noise sources. For example, the adaptive directional microphone should adopt the bidirectional pattern if the dominant noise source is located at the 90° or 270° azimuths and adopt the cardioid pattern if the dominant noise source is located at 180° azimuth. In practice, different hearing aid manufacturers use different calculation methods to estimate the location of the dominant noise source and to vary the internal delay of the directional microphones accordingly. The actual location of the null may vary, depending on the calculation method and the existence of other noise and sounds in the environment.

The adaptive ability of the adaptive directional microphones is achieved in three to four steps:

Table 1 summarizes the characteristics and strategies implemented in adaptive directional microphones of some hearing aids. Notice that the determination of the operational mode is user-determined for GNReSound Canta but automatic for other hearing aids. Another point worth noting is that most adaptive directional microphone algorithms process the signal in a single band. More recently, multichannel adaptive directional hearing aids have been introduced. First introduced in the Oticon Syncro hearing aids, this technology allows different directional sensitivity patterns to occur within multiple channels at the same time.

OPEN IN VIEWER

Table 1.

The Characteristics of Adaptive Directional Microphones Implemented in Selected Commercially Available Hearing Aids*

	Oticon-Syncro	Phonak-Perseo	ReSound-Canta	Siemens-Triano	Widex-Diva
Signal Detection and Analysis	Global modulation detector to determine the continuous signal to noise ratio. Parallel processing to calculate the resultant signal-to-noise level in each possible microphone mode and polar configuration Level detector to determine the overall sound pressure level of the incoming signal Front-back ratio detector to estimate the location of the dominant input signal Wind noise detector to estimate the level of wind noise	Level detector to determine the overall sound pressure level of the incoming signal Front-back ratio detector to estimate the location of the dominant input signal Analysis of the amplitude modulations, temporal fluctuation, and spectral center of gravity of the incoming signal to infer the presence of speech and noise	Front-back ratio detector to estimate the location of the dominant sound source	Level detector to determine the overall sound pressure level of the incoming signal Wind noise detector to detect the presence and the level of wind noise	Front-back ratio detector to estimate the location of the dominant input signal Noise classification to determine noise type: wind, circuit or environmental Level detector to determine the overall sound pressure level of the incoming signal
Decision Rules for Determining the Microphone Mode	Surround Mode: If the omni-directional mode in all 4 bands provides the highest SNR If the incoming signal is at soft to moderate levels with no or low background noise If the dominant speaker is from the back If strong wind is detected Split-Directionality Mode: If the omni-directional mode in the lowest frequency band and directional mode at the upper 3 bands provide the highest SNR If the incoming signal is at a moderate level with some background noise If a moderate level of wind noise is detected Full-Directionality Mode: If the directional mode in all 4 bands gives the highest SNR If the incoming signal is high with background noise If no wind noise is detected	Omni Mode: speech only Directional Mode: The decision rules for switching to directional microphones can be adjusted by the clinician on the basis of user priority for speech audibility or comfort: Audibility: the algorithm switches to directional microphone mode when speech in noise is detected, but not when speech alone or noise alone is detected Comfort: the algorithm switches to directional microphone mode whenever there is noise, regardless of whether speech is present or not. If only speech is present, it is not activated	User determined	Omni Mode: Analysis indicates that primary signal is speech or wind noise The intensity of incoming signal is below a predetermined level (e.g., 60 dB SPL). The trigger level varies depending on the hearing aid model. Directional Mode: Minimal wind noise is detected Level of the incoming signal exceeds the predetermined level	Omni Mode: If incoming signal is from the front only If environmental noise is insignificant If wind noise is the dominating noise source Directional Mode: Minimal wind noise is detected Level of the incoming signal exceeds the predetermined level Background noise is detected
Adaptation Speed for Omni-Directional and Directional Switch	2–4 sec, depending on the hearing aid's Identity setting, i.e., the life style of the hearing aid user in the fitting software	Variable/programmable by clinician, from 4–10 sec, based on “Audibility” or “Comfort” selections in the hearing aid fitting software	Not applicable because switch is user determined	6–12 sec, depending on the settings of the listening program	5–10 sec, depending on the settings of the listening program
Decision Rules for Determining the Polar Patterns	The analyzer unit calculates the signal-to-noise level at each azimuth from 90° to 270° for each polar pattern across the four bands. The internal delays of the polar patterns that generates the best SNRs are adopted	The internal delay that yield the minimum power output from the directional microphone is adopted	The internal delay that yield minimum output from the directional microphone is adopted	The weighted sum of a bidirectional and cardioid pattern is calculated and the internal delay that yields the minimum output (weighted sum) from the directional microphone is adopted	The analyzer unit receives the sounds from the front and back omnidirectional microphones and adopts the internal delay that would give the lowest output at the directional microphone output
All hearing aids: Any polar pattern with nulls between 90° to 270° is possible
Adaptation Speech Between Different Polar Patterns	2 sec/90°, speed may vary depending on the hearing aid's Identity setting	100 ms between polar patterns	Analysis of environment every 4 ms, changing of polar pattern every 10 ms	50 ms/90°	Typically less than 5 sec
Polar Pattern when Multiple Noise Sources Exist	Cardioid in uniform noise field Multiband directionality enables different polar patterns to reduce the level of the multiple noise sources if they have different frequency contents	Cardiod	Hypercardioid	Hypercardiod	Hypercardioid
Low Frequency Equalization	Automatic	Programmable in fitting software via “Contrast” feature	Programmable in fitting software	Automatic	Automatic for each polar patterns
Information Source(s)	Oticon, 2004; Flynn, 2004, personal communication	www.Phonak.com (a); Ricketts and Henry (2002); Fabry (2004), personal communication	Groth (2004), personal communication	Powers (2004), personal communication. Powers & Hamacher (2004)	Kuk et al., 2002a; Kuk, 2004, personal communication
Clinical Verification	Flynn (2004): compared to the first-order fixed directional microphone implemented in Adapto, Syncro's Full-Directionality mode combined with its noise reduction algorithm yielded about 1–2 dB better SNR-50s for hearing aid users with multiple broadband noise sources in the back hemisphere. It is unclear how much of the improvement is solely generated by the adaptive directional microphone	Unavailable. See text for the evaluation of the first-order adaptive directional microphone implemented in Phonak Claro	Unavailable	Bentler et al. (2004a): the hybrid second-order adaptive directional microphone has improved the SNR-50s of hearing aid users for 4 dB. No significant difference in SNR-50s between the first-order and the hybrid second-order adaptive directional microphones. Ricketts et al. (2003): Significant benefit was observed using the second-order directional microphone compared to its fixed directionality mode in moving noise	Valente and Mispagel (2004): Compared to the omni-directional mode, the adaptive directional microphone improved SNR-50s for 7.2 dB if a single noise source was located at 180°. The improvement in SNR-50s decreased to 5.1 dB and 4.5 dB when noise was presented at 90° + 270°, and 90°+180°+270°, respectively

*

These hearing aids are selected to demonstrate the range and the differences in implementation methods of adaptive directional microphone algorithms in commercially available hearing aids. SNR = signal-to-noise ratio.

It is apparent in Table 1 that hearing aid manufacturers use different strategies to implement their adaptive directional microphone algorithms. The following discussion explains the similarities and differences among the adaptive directional microphone algorithms from different hearing aid manufacturers or models.

a. Signal Detection and Analysis

In the signal detection and analysis unit, algorithms implemented in different hearing aids may have a different number of signal detectors to analyze different aspects of the incoming signal. Some of the most common detectors are the level detector, modulation detector, spectral content analyzer, wind noise detector, and front-back ratio detector, among others.

i. Level Detector

The level detector in adaptive directional microphone algorithms estimates the level of the incoming signal. Many adaptive directional microphones only switch to directional mode when the level of the signal exceeds a certain predetermined level. At levels lower than the predetermined level, the algorithm assumes that the hearing aid user is in a quiet environment and the directional microphone is not needed. Thus, the hearing aid stays at the omni-directional mode.

ii. Modulation Detector

The modulation detector is commonly used in hearing aids to infer the presence or absence of speech sounds in the incoming signal and to estimate the SNR of the incoming signal. The rationale is that the amplitude of speech has a constantly varying envelope with a modulation rate between 2 and 50 Hz (Rosen, 1992), with a center modulation rate of 4 to 6 Hz (Houtgast and Steeneken, 1985). Noise, on the other hand, usually has a modulation rate outside of this range (e.g., ocean waves at the beach may have a modulation rate of around 0.5 Hz) or it occurs at a relatively steady or unmodulated level (e.g., fan noise).

The speech modulation rate of 4 to 6 Hz is associated with the closing and opening of the vocal tract and the mouth. Speech in quiet may have a modulation depth of more than 30 dB, which reflects the difference between the softest consonant (i.e., voiceless /ø/) and the loudest vowel (i.e., /u/). Modulation depth is the level difference between the peaks and toughs of a waveform plotted in the amplitude-time domain (Figure 5). If a competing signal (noise or speech babble) is present in the incoming signal, the modulation depth decreases. Because the amount of amplitude modulation normally decreases with an increase in noise level, the signal detection and analysis unit uses the modulation depth of signals with modulation rates centered at 4 to 6 Hz to estimate the SNR in the incoming signal—the greater the modulation depth, the higher the SNR. Notice that if the competing signal (noise) is a single talker or has a modulation rate close to that of speech, the signal detection and analysis unit cannot differentiate between the desired speech and the noise.

OPEN IN VIEWER

Figure 5.

The modulation detector is composed of a maxima (thick line) and a minima follower (thin line). The maxima follower estimates the level of speech and the minima follower estimates the level of noise. The difference between the two allows the estimation of signal-to-noise ratio in the frequency channel.

The modulation detector is used in the adaptive directional microphone algorithms of the Oticon Syncro and Phonak Perseo digital hearing aids. However, the results of the modulation detectors are used to make different decisions in the algorithm.

The modulation detector of Perseo analyzes the modulation pattern and the spectral center of gravity to estimate the presence or absence of speech and noise. An analog to the spectral center of gravity is the center of gravity for an object. The difference is that center of gravity refers to the weight center of the object, whereas spectral center of gravity refers to the frequency center of a sound. The result of the modulation detector in Perseo is then combined with the priority setting (i.e., Audibility or Comfort) and used to determine the appropriate operational mode for the instance.

Syncro, on the other hand, uses the results of the modulation detector to calculate SNR at the output of the directional microphone. The signal processing algorithm is programmed to seek the operational mode (i.e., the Surround Mode, Split-or Full-Directionality Modes) and the polar patterns to maximize the SNRs of the four frequency bands at the microphone output (Flynn, 2004a). Syncro defines the speech as waves with modulation rates ranged from 2–20 Hz.

iii. Wind Noise Detector

The wind noise detector is used to detect the presence and the level of wind noise. Although the exact mechanisms used in the algorithms from different manufacturers are unknown, it is possible that the wind noise detectors make use of several physical characteristics of the wind noise and hearing aid microphones to infer the presence or absence of wind. First, directional microphones are more sensitive to sounds coming from the near field than sounds coming from the far field, whereas omni-directional microphones have similar sensitivity to sounds coming from the near field and the far field. To a dual-microphone directional microphone, near field refers to sounds coming from a distance of within 10 times of the microphone spacing; far field refers to sounds coming from a distance of more than 100 times of the microphone spacing. Sounds coming from a distance of between 10 to 100 times of the microphone spacing have the properties of both a near field and a far field (Thompson, 2004, personal communication).

When a sound comes from the far field, the outputs generated at the two omni-directional microphones that form the directional microphone are highly correlated. If the outputs are correlated 100%, the peaks and valleys of the waveform from the two microphone outputs should coincide when an appropriate amount of delay is applied to one of the microphone output during the cross-correlation process. The amount of delay applied depends on the direction of the sound. In other words, the outputs of the two omni-directional microphones have a constant relationship and similar amplitude for a sound coming from the far field.

Because microphone output is highly correlated for sounds from the far field, when the microphone outputs are delayed and subtracted to form a directional microphone, the amplitude of the signal is reduced if the signal comes from the sides or the back hemisphere and not much affected if the signal comes from the front hemisphere. In addition, the directional microphone exhibits a 6-dB/octave roll-off at the low-frequency region for sounds coming from any direction. Assuming that the frequency response of the hearing aid is compensated for low-frequency roll-off, the output of the omni-directional microphone mode is comparable to the output of the directional microphone mode for sounds coming from the far field (Edwards, 2004, personal communication; Thompson, 2004, personal communication).

When wind is blowing, a turbulence and some eddies are generated close to the head. Wind noise is therefore a sound from the near field. For a sound coming from the near field, the outputs of the two omni-directional microphones that form a directional microphone are poorly correlated. When the outputs of the omni-directional microphones are delayed and subtracted, minimum reduction in amplitude results no matter which direction the sounds are coming from. In fact, the wind noise entering the two microphones are added to further increase the sensitivity of the directional microphone to wind noise, especially at high frequencies. In addition, the output of the directional microphone also does not exhibit a 6-dB/octave roll-off at the low-frequency region; that is, the frequency response of the sounds is similar for the directional and the omni-directional modes. Assume that it is the same hearing aid with low-frequency compensation; now, the output of the directional microphone is much higher than the output of the omni-directional microphone for this near field sound because of the increased sensitivity and the low frequency gain compensation (Edwards, 2004, personal communication; Thompson, 2004, personal communication).

Although the exact mechanisms of wind noise detectors are proprietary to each hearing aid manufacturer, it is possible that one characteristic that the wind noise detector monitors is the differences between the outputs of the omni-directional and directional microphones (Edwards, 2004, personal communication). Using the example with equalized low-frequency gain, the outputs of the omni-directional and directional microphones are comparable for sounds coming from the far field, but the output of the directional microphone is much higher than the omnidirectional microphone for sounds coming from the near field (wind noise). On the other hand, if the low-frequency gain is not equalized, the output of the directional microphone is lower than the output of the omni-directional microphone for sounds coming from the far field, but the output of the directional microphone is higher than the output of the omni-directional microphone for sounds from the near field.

Another possible strategy to detect wind noise is to use the correlation coefficient to infer the presence of wind noise. The correlation coefficient can be determined by applying several delays to the output of one of the omni-directional microphones and calculating the correlation coefficient between the outputs of the two microphones for each delay time. As mentioned previously, if the microphone outputs are correlated 100%, the peaks and valleys of the waveforms coincide perfectly. If the peaks and valleys of the waveform are slightly mismatched in amplitude or phase, the outputs are said to have a lower correlation coefficient. For sounds in the near field, the correlation coefficient can be close to 0%.

The wind noise detector can make inference based on the degree of correlation between the outputs of the two omni-directional microphones. If the outputs have a high correlation coefficient, the wind noise detector infers that wind noise is absent. If the outputs have a low correlation coefficient, the algorithm infers that wind noise is present (Thompson, 2004, personal communication; Siemens Audiology Group, 2004). According to Oticon, the wind noise detectors in the Syncro hearing aids detect the uncorrelated signals between the microphone outputs that are consistent with the spectral pattern of wind noise to infer the presence or absence of wind noise (Flynn, 2004, personal communication).

In addition, it is possible that a wind noise detector can set different correlation criteria for the coefficients at low- and high-frequency regions for wind noise reduction. High-frequency eddies are normally generated by finer structures around the head (e.g., pinna, battery door of an in-the-ear hearing aid) and low-frequency eddies are generated by larger structures (e.g., the head and the shoulders). As the finer structures are much closer to the hearing aid microphones (in the near field) and the larger structures are further away from the microphone (in the mixed field), high-frequency sounds tend to have a lower correlation coefficient than low-frequency sounds at the microphone output (Thompson, 2004, personal communication). A sample decision rule for the wind noise detector to make use of this acoustic phenomenon can be: wind noise is present in the microphone output if the correlation coefficient is less than 20% at the low-frequency region and less than 35% at the high-frequency region.

When wind noise is detected, many hearing aids with adaptive directional microphones either remain at or switch to the omni-directional microphone mode to reduce annoyance of the wind noise or to increase the audibility of speech, or both (Kuk et al., 2002b; Oticon, 2004a, Siemens Audiology Group, 2004).

iv. Front-Back Detector

Some adaptive directional microphone algorithms also have a front-back ratio detector that detects the level differences between the front and back microphones and estimates the location of dominant signals (Fabry, 2004, personal communication, Groth, 2004, personal communication, Kuk, 2004, personal communication; Oticon, 2004a). For example, the front-back detector of Oticon Syncro combines the analysis results of the front-back ratio detector and the modulation detector to determine if the dominant speech is located at the back. If a higher modulation depth is detected at the output of the back microphone, the algorithm would remain at or switch to the omni-directional mode (Oticon, 2004a).

3.1.2.2. Verification and Limitations

3.1.3.1. Verification and Limitations

Bentler and colleagues (2004a) measured a random sample of behind-the-ear Triano hearing aids with the hybrid second-order directional microphones and reported the free-field average directivity index (DI-a) values ranged from 6.5 to 7.8 dB. The DI-a values were calculated from the sum-average of the DI values from 500 to 5000 Hz without frequency weighting. When the Triano hearing aids were worn on a Knowles Electronics Manikin for Acoustic Research (KEMAR), the DI-a values ranged from 4.5 to 6.0 dB.

Several research studies have investigated the effectiveness of the hybrid second-order directional microphones in stationary and moving noises. Bentler and colleagues (2004a) compared the speech understanding performance of subjects with normal hearing and subjects with an average of 30 to 65 dB HL hearing loss from 250 to 8000 Hz. Subjects with normal hearing listened to the Hearing in Noise Test (HINT, Nilsson et al., 1994) in stationary and moving noises to serve as the standard. Subjects with hearing loss were fit with a pair of Triano hearing aids and a pair of another hearing aid with a first-order adaptive directional microphone by the same manufacturer. They listened to the HINT sentences in omni-directional and directional modes in a stationary noise field and in omni-directional, directional, and adaptive directional modes in a moving noise field.

The results indicated that subjects with hearing loss exhibited an aided SNR-loss of 4 dB in stationary noise and slightly less than 5 dB in a moving noise. The performance of subjects with hearing loss showed roughly a 4 dB of SNR improvement in stationary noise by using Triano compared with the hearing aids with first-order directional microphones. Subjects with hearing loss also obtained a little more than 3 dB of SNR improvement using the hearing aids with first-order adaptive directional microphone and approximately 4 dB of improvements using Triano adaptive directional microphone. No significant differences were found in subjects’ performance between Triano and the hearing aid with first-order directional microphones in either noise field.

Ricketts and colleagues (2003) have reported that the hybrid second-order adaptive directional microphones yielded approximately 2 dB lower SNR-50 than the same hearing aid at the fixed directional microphone mode in the presence of a moving noise source at the back hemisphere. They also reported that the fixed and adaptive directional microphones generated about 5.7 dB and 7.6 dB lower SNRs-50 than the omni-directional mode of the hearing aid.

3.1.4.1. Verifications and Limitations

A properly functioning microphone matching algorithm should provide individualized in situ matching of the microphones and maximize the directional performance of the directional microphone throughout the lifetime of the hearing aid. The rationale of microphone matching algorithms sounds very logical and promising; however, the exact procedures used by different manufacturers are unknown to the public. To date, no verification data on the effectiveness of these algorithms are available.

The limitation of the microphone matching algorithm is that it cannot protect against factors (e.g., clogged microphone ports) other than the microphone drift. In fact, Thompson (2003) argued that the degradation in directional performance of the directional microphones is often due to the condensation of debris on the screen or clogged microphone ports rather than microphone drift. As reasons other than microphone mismatch may determine the performance of the directional microphones over time, constant monitoring of the microphones’ physical conditions is very important. Clinicians need to check the conditions of the two omni-directional microphones under a microscope or amplifying lens during regular clinic visits to ensure that the microphone openings are free of debris and the microphone screens are clearly seen and well defined, even though the hearing aids are equipped with microphone-matching algorithms (Ricketts, 2001).

a. Head-Worn Array Microphones

Head-worn array microphones can be implemented as either an end-fire or a broadside array. An end-fire array has its most sensitive beam parallel to the microphone array, such as an array microphone implemented along an arm of a pair of eyeglasses. A broadside array has its most sensitive beam perpendicular to the microphone array, such as an array microphone implemented above the glasses of a pair of eyeglasses.

Etymotic Research designed and marketed an end-fire array microphone, Link.It. Link.It uses delay-and-sum processing to combine the outputs of three single-microphone directional microphones. Each directional microphone is spaced 25 mm apart, and the outputs of the second and third directional microphones are delayed for 75 and 150 milliseconds, respectively. When sounds come from the front, the outputs of the microphones are in phase after accounting for the traveling time and the delay circuit. The sum of the outputs from the three microphones is three times as large as the single directional microphone output.

When sounds come from the sides, the outputs of the three microphones are 180° out of phase because of the delay added to the output of the microphones (Christensen et al., 2002). According to Etymotic Reseach (http://www.etymotic.com/ha/linkit-ts.asp), the single-microphone directional microphones (instead of omnidirectional microphones) are used to optimize the performance of the array microphone over time and minimize the need to monitor and match the sensitivity and the phase of the microphones.

Link.It sends its processed signal to the hearing aid wirelessly via telecoil (Figure 8A). If necessary, the output of Link.It can be fed into the direct audio input of the hearing aid. Link.It has a relatively flat frequency response from 200 Hz to 4000 Hz. When measured on KEMAR in an anechoic chamber, it yielded an AI-DI of 7 dB on KEMAR and 8 dB in free-field (Christensen et al., 2002).

OPEN IN VIEWER

Figure 8.

(A) Link.It transmits the processed signal to an in-the-ear hearing aid via telecoil. (Courtesy of Etymotic Research, reprinted with permission). (B) The hand-held unit of Lexis transits the processed signal to an frequency modulated (FM) receiver attached to a behind-the-ear hearing aid. (Courtesy of Oticon, reprinted with permission).

b. Hand-Held Array Microphones

Most hand-held array microphones are implemented in the end-fire array (i.e., shot-gun microphone array). Recently, a new hand-held array microphone, Lexis, has been introduced. Lexis has a hand-held unit with an array microphone and a built-in FM transmitter. Signals from Lexis are sent to the hearing aid via an FM receiver plugged into the direct audio input of the hearing aid.

The hand-held unit of Lexis is composed of four single-microphone directional microphones aligned on the side of the unit (Figure 8B). Again, single-microphone directional microphones are used to maintain the directional effect over time while minimizing the need to monitor and match the sensitivity and phase of the microphone components. The port spacing between these single-microphone directional microphones is 15 mm. According to Oticon (2004b), 15 mm was chosen as a compromise between the amount of low-frequency roll-off (the larger the port spacing, the less the low-frequency roll-off) and high directivity at the high-frequency range (the smaller the port spacing, the higher the high-frequency directivity index).

Lexis has three user-switchable directionality modes: omni-directional, focus, and superfocus. The superfocus mode has a narrower sensitive beam to the front than the focus mode. The AI-DI at the superfocus mode is reported to be 8.5 dB and 5.9 dB in the focus mode. Lexis has a relatively flat frequency response, from 600 Hz to 5000 Hz (Oticon, 2004b). During one-on-one communication or listening, the hand-held unit can be worn around the neck of the talker like the microphone and transmitter unit in other FM systems.

3.1.5.1. Verifications and Limitations

Clinical trials of Link.It (Christensen et al., 2002) and studies carried out during its developmental stages (Bilsen et al., 1993; Soede et al., 1993) reported a 7 to 10 dB SNR improvement for people with hearing impairment in noisy, reverberate environments.

Oticon (2004b) conducted a “just-follow-the-conversation” test in a laboratory setting. During the test, speech was fixed at 65 dB SPL and subjects were asked to adjust the level of the noise so that they could understand 50% of the information. The results indicated that five subjects with moderate-to-profound hearing loss obtained 5.6 and 8.7 dB of directional benefit for the focus and superfocus modes relative to the omni-directional mode, respectively. An interesting observation is that when Lexis is used in the hand-held position, the omni-directional mode of Lexis is about 4 dB better than the omni-directional mode of the subjects’ own hearing aids because of the body baffle effect. Significant improvement was also reported in all subtests of the Abbreviated Profile of Hearing Aid Benefit (APHAB, Cox and Alexander 1995) when the subjects’ own hearing aids were compared with the superfocus mode of Lexis.

Super-directionality is a double-edged sword. On one hand, it has very high sensitivity to sounds from a very narrow beam to the front and it can reduce background noise significantly. This feature is especially useful for listening to a talker located at a fixed direction or a talker moving at a predictable path. On the other hand, if several talkers are participating in a discussion or conversation, say in a round table, it may be extremely hard for the user to zoom in to the correct talker as the beam is so narrow. If the talker is not in the beam of the directional microphone sensitivity, the user has to rely on visual cues to locate the talker and then point the hand-held unit to the talker. The user may miss the first several words whenever the talker changes. In such a case, the focus mode may be more appropriate because it has a wider sensitivity beam than the superfocus mode. The drawback is that it is less directional and thus the noise reduction ability is less than that of the superfocus mode. Another caution when using Lexis is that highly directional devices may reduce the user's ability to hear warning sounds from locations, such as the sides, with low microphone sensitivity.

It is worthy of noting that although array microphones can provide up to 7 to 8 dB of improvement in SNR, FM systems have been shown to provide 10 to 20 dB of improvement (Crandell and Smaldino, 2001; Lewis et al., 2004). FM systems can remarkably improve the SNR because the microphone is usually located near the mouth of the talker, thus they significantly reduce the effects of reverberation, distance and noise. Therefore, in situations where the voice of one talker is desirable (e.g., one-on-one conversation in classrooms or lecture halls), the use of FM systems or array microphones configured to function as FM systems (i.e., the hand-held unit of Lexis worn around the talkers’ neck) is recommended.

3.2.1.1. How They Work

b. Decision Rules

The decision rules of a noise reduction algorithm may depend on several factors. The most common of these include the estimated modulation depth/SNR, frequency importance function, the level of the incoming signal, and the degree of noise reduction selected in the hearing aid fitting software. The amount of gain reduction in each channel is usually inversely proportional to the SNR estimated in the frequency channel (Kuk et al., 2002b; Powers and Hamacher, 2002; Johns et al., 2002; Edwards et al., 1998; Latzel, Kiessling and Margolf-Hackl, 2003; Schum, 2003; Walden et al., 2000).

This approach is based on the rationale that if the signal-detection and analysis unit estimated a high SNR in a frequency channel, the algorithm assumes that speech-in-quiet is detected in the frequency channel and the action unit should let the signal pass without attenuation. If the unit estimates a moderate or low SNR in the frequency channel, the algorithm assumes that either speech coexists with noise or noise dominates the frequency channel. Thus, the gain of the frequency channel should be reduced to decrease the noise interference. When no modulation is detected in a frequency channel, the analysis unit assumes that no speech is present in the frequency channel and maximum attenuation should be applied.

The modulation depth at which the gain reduction starts to be applied at a frequency channel is sometimes called the “modulation threshold for noise reduction activation” (Groth, 2004, personal communication). The modulation threshold for noise reduction activation and the exact amount of gain reduction applied at different SNRs differs among the noise reduction algorithms. Figure 9 illustrates the relationship between the estimated SNR and the amount of gain reduction in two commercially available digital hearing aids.

OPEN IN VIEWER

Figure 9.

The relationship between the estimated signal-to-noise ratio (SNR) and the amount of gain reduction applied by noise reduction algorithms of two commercial digital hearing aids.

Another common consideration in the decision rules of the multichannel adaptive noise reduction algorithms is the frequency-importance weighting of the frequency region for speech understanding. One of the approaches is to set the amount of gain reduction inversely proportional to the articulation index of the frequency region (Kuk et al., 2002b; Alcantara et al., 2003; Boysmans and Dreschler, 2000). The assumption is that as the weightings of the articulation index increase, the importance of the frequency channel for speech understanding also increases; therefore, less gain reduction should be applied to these frequency channels (Kuk et al., 2002b; Oticon, 2004a).

Other manufacturers may also use different sets of gain reduction rules to account for the importance of speech information in the frequency channel (Alcantara et al., 2003; Tellier et al., 2003). For example, Phonak Claro only reduces the gain at frequency channels below 1 kHz and above 2 kHz. The rationale is that frequencies between 1 and 2 kHz are very important for speech understanding; therefore, the gain is not reduced regardless of the modulation depth of the incoming signal at those frequencies (Alcantara et al., 2003). Another form of frequency-dependent gain reduction is found in Siemens’ Prisma, in which the amount of maximum gain reduction in a frequency channel can be programmed by the clinician in the hearing aid fitting software (Powers et al., 1999).

In addition to the modulation depth and the importance of speech content in the frequency channel, some manufacturers may add another dimension to their gain reduction decision rules: the sound pressure level of the incoming signal or the sound pressure level of the noise. For example, if a particular modulation depth is detected within a frequency channel, the Widex Diva starts to reduce the gain of the frequency channel only if the input level exceeds 50 to 60 dB. The amount of reduction also increases as the level of the incoming signal increases. If the modulation depth is high and the level is low, no gain reduction is applied (Kuk et al., 2002b). The assumptions are that noise reduction is not needed in quiet or in environments with low levels of noise and a higher amount of noise reduction is needed in a noisier environment.

Many multichannel adaptive noise reduction algorithms allow the clinician to choose the degree of noise reduction in the fitting software. As the degree of noise reduction increases, the maximum gain reduction also increases. This maximum gain reduction is usually applied across frequency channels without affecting the frequency weighting of the particular channel (Tellier et al., 2003).

3.2.1.2. Verification and Limitations

b. Interaction Between Multichannel Adaptive Noise Reduction Algorithm and Wide Dynamic Range Compression

A caution for fitting hearing aids with wide dynamic range compression and noise reduction is that wide dynamic range compression may reduce the effectiveness of noise reduction algorithms. The interactions between wide dynamic range compression and noise reduction algorithm can be seen when these two signal processing units are implemented in series. Specifically, interactions exist if the level detector of the compression system uses the output of the noise reduction algorithm to make decisions on the amount of gain that is applied to the signal or if the noise reduction algorithm uses the output of the compression system to make decisions of noise level and modulation.

To illustrate the interactions, Figure 10A displays the amplitude envelope of two sentences presented to a diffuse sound field at SNR of 3 dB. Figure 10B and C are the same sentences processed by a digital hearing aid with its noise reduction algorithm activated and the compression system programmed to linear and 3:1 compression, respectively. The compression system has fast time constants and was implemented in series with the noise reduction algorithm in the signal processing path. The frequency responses in the compression and linear modes were matched at the presentation level. Compared with the envelope of the sentence processed in the linear mode, the envelope of the sentence processed in the compression mode exhibits a lower modulation amplitude. The amplitude of the noise between the sentences was also higher in the compression mode than the linear mode. The combination of low modulation depth and higher noise level suggests that a 3:1 compression reduced the modulation depth and thus the SNR of the processed signal. The reason for the increase in noise level is because wide dynamic range compression provides more gain for soft sounds (i.e., noise in this case) and less gain for loud sounds (i.e., speech in this case).

OPEN IN VIEWER

Figure 10.

The interaction between wide dynamic range compression and the noise reduction algorithm when the two systems are implemented in series. (A) The amplitude envelope of two sentences and speech spectrum noise presented in sound field at a signal-to-noise ratio (SNR) of +3 dB. (B). The amplitude envelope of the same sentences and noise after being processed by a directional microphone and a noise reduction algorithm with the compression system set in the linear mode. (C). The envelope of the same sentences and noise after being processed by a directional microphone and a noise reduction algorithm with the wide dynamic range compression system set at 3:1 compression. The frequency responses of the hearing aid in the linear and compression mode were matched at the presentation level.

To date, few research studies have investigated the perceptual effects of the interactions between wide dynamic range compression and noise reduction algorithms on speech understanding and sound quality. Research studies are also needed to explore if any interaction exists between noise reduction algorithms and compression systems when the two systems are implemented in parallel in the signal processing path.

An example of the parallel implementation of the noise reduction algorithm and the compression system is that the signal detectors of the noise reduction algorithm and the compression system detect and make decisions based on the signal at the microphone output. Clinicians need to keep in mind that the interactions between the noise reduction algorithm and the wide dynamic range compression system with a high compression ratio might reduce the effectiveness of noise reduction algorithms.

3.2.2.1. How It Works

The synchrony detection noise reduction algorithm makes use of the co-modulated/synchronous nature of the speech sounds to detect the presence of speech (Elberling, 2002; Schum, 2003). The signal detection unit of the noise reduction algorithm constantly monitors the incoming signal at high-frequency bands (i.e., upper three bands in Adapto) for synchronous energy at the rate the fundamental frequencies of human voices. According to Oticon, the signal detection unit is capable of detecting synchronous energy down to −4 dB SNRs (Flynn, 2004, personal communication).

The synchrony detection noise reduction algorithm is implemented in Oticon Adapto hearing aids. If synchronous energy in the upper three frequency bands is not detected, the noise reduction algorithm assumes that no speech signal is present and the noise reduction unit gradually reduces the overall gain by decreasing the gain at high input levels (i.e., the compression ratio is increased) at all frequency bands. When synchronous energy is detected, the hearing aid returns to normal settings instantaneously and allows the signal to pass without attenuation. In other words, the detection of synchronous energy across the frequency bands deactivates the actions of the noise reduction algorithm (Schum, 2003; Bachler et al., 1995; Elberling, 2002).

3.2.2.2. Verifications and Limitations

The synchrony-detection noise reduction algorithm is designed to take advantage of the temporal separation between speech and noise because it acts at the instances when speech is not present and allows the signal to pass when speech is present. The goal of this algorithm is to increase listening comfort in the absence of speech signals. Yet it does not provide any benefit in listening comfort or speech understanding when speech and noise coexist or when speech is the competing signal. The synchrony-detection noise reduction algorithm is solely implemented in hearing aids by Oticon. No validation data is available on its effectiveness.

3.2.3.1. How it Works

The noise reduction algorithm of Syncro has three detectors in the signal detection and analysis unit: a synchrony detector, a modulation detector, and a level detector (Table 2). The synchrony detector monitors the presence or the absence of synchronous energy across the upper four frequency channels to infer the presence or absence of speech in the incoming signal. The modulation detector monitors the modulation depth and the noise level of the incoming signal within each frequency channel. The noise level detector determines the noise level in the incoming signal.

The Syncro Optimization Equation integrates the information from these detectors and determines if the incoming signal has speech only, speech with noise, or noise only. Then it examines the current instrument settings, calculates the output of each of the three states, and then decides the amount of gain reduction that should be applied to each frequency channel to maximize the SNR for that particular instance (Table 2). In general, the amount of gain reduction that should be applied to each frequency channel depends on the modulation depth, the noise level, and the weighting of articulation index of the frequency channel (Figure 12) (Flynn, 2004a, Oticon, 2004a).

OPEN IN VIEWER

Figure 12.

The decision rules of Oticon Syncro. The synchrony detector, the modulation detector and the noise level detector determine the presence or the absence of speech and the relative level of speech and noise in the incoming signal. No gain reduction is applied to the frequency channels if speech-in-quiet is detected. If speech-in-noise is detected, the gain is reduced depending on the level of the noise, the modulation depth, and the articulation index weighting of the frequency channel. Maximum gain reduction is applied if noise-only is detected. (Reprinted with permission from Oticon The Syncro Concept, 2004).

3.2.3.2. Verification and Limitations

Syncro was recently introduced into the hearing aid market. No verification data on the effectiveness of its noise reduction algorithms are available.