Despite the rapid improvement in hearing aid technology, the performance of hearing aids in noise remains a main reason for wearer dissatisfaction. One of the many reasons for nonuniform wearer satisfaction in noise is that the design and implementation of noise management strategies (to include noise reduction algorithms and directional microphones) is not uniform among manufacturers.
This article examines some of the desirable considerations to illustrate more effective methods of noise management.
Criteria for Noise Management Algorithms
Improved performance in noise… Obviously, the objective of any noise management strategy is to ensure wearer satisfaction in noise. But the criteria for satisfaction vary across listeners. While most would accept improved speech understanding in noise to be a criterion for satisfaction, many would accept improved listening comfort or increased tolerance in noise as meeting their criterion.
Indeed, studies on the current single microphone noise reduction algorithms generally conclude that such algorithms do not improve speech understanding in noise. Rather, improved listening comfort and increased acceptance of the hearing aid in noise are reported as the general benefit of single microphone noise management algorithms.1
On the other hand, the use of directional microphones as a noise management strategy has led to greater success. Such a system has reported 2 to 3 dB improvement in signal-to-noise ratios (SNR) in an open fitting2 and 4 to 6 dB improvement in SNR in a more occluded fitting.3 The improvement is seen in all degrees of hearing losses, but the magnitude of the improvement is contingent upon the angle of stimulus presentation.4 Greater SNR improvement is found where the “speech” signal separates from the interfering noise stimulus.
…while preserving audibility and intelligibility? The “do no harm” principle should also be followed in the design of noise management systems. That is, the design should not introduce artifacts and degrade intelligibility. Degradation is a potential because noise management strategies (both noise reduction algorithms and directional microphone systems) achieve their intended functions by “reduction.”
In the case of a noise reduction algorithm, gain reduction occurs when a “noise” decision is made of the incoming stimulus.5 In the case of a directional microphone, the sensitivity of the microphone is reduced to sounds from a specific direction; the input to the hearing aid amplifier is reduced.6
It is foreseeable that, if gain is reduced inappropriately or when the desirable sounds (such as speech) originate from the sides and back, intelligibility could suffer.7 Thus, a major criterion in judging a noise management strategy should also include how it preserves speech intelligibility.
Preserving Speech Intelligibility: Noise Management
To ensure maximum intelligibility, the following principles should be considered for each noise management system.
1) Accurate identification of “noise” versus “non-noise.” One way to ensure audibility is to avoid unnecessary gain reduction. This can be achieved with a mechanism that can accurately identify “noise” (versus “non-noise”). If gain reduction occurs only for “noise” or unwanted signals, it would avoid gain reduction for the “non-noise” or “speech” signals. A level distribution analysis is one approach to distinguish between “noise” and “non-noise.”5 In situations where both speech and noise are present and for signals that are “speech-like” but are unwanted by the wearers, additional safeguards are implemented to minimize noise but still preserve intelligibility.
2) Appropriate activation threshold and gain reduction. Most people object to noise when its level is loud. It makes extended listening uncomfortable or intolerable (circuit noise is a different issue and can be solved without the typical noise reduction algorithm). Consequently, one way to ensure audibility is to reduce gain only for loud “noise” but not for soft “noise.” Doing so has the advantage that, even if the identification of “noise” and “speech” is inaccurate (ie, identifying speech as noise), reducing gain for loud sounds only makes the signal softer, but not completely inaudible. It would not be the case when gain for soft sounds is reduced.
Thus, the activation threshold of a noise reduction algorithm should be above 60 dB SPL in order to preserve the audibility of soft sounds. Furthermore, only the gain parameter for loud sounds should be affected. This ensures audibility for the softer sounds.
3) Differential gain reduction. Gain reduction affects speech and noise signals equally if both are present at the same time. In order to minimize the negative impact on speech intelligibility, for the same SNR, less gain reduction may be applied to the frequencies that are more important for speech intelligibility.
These considerations have preserved intelligibility with the use of noise reduction algorithms,8 but they alone may not preserve audibility of noise reduction algorithms for people with a severe-to-profound hearing loss. This is because the action of the current noise reduction system is based solely on the characteristics of the input signals. The higher the input level, or the poorer the SNR, the more gain reduction one may expect from the noise reduction algorithm. Typically, a maximum of 10 to 12 dB of gain reduction may be expected.
For someone with a mild-to-moderate degree of hearing loss, a gain reduction of 10 to 12 dB may still leave the amplified sound audible. On the other hand, reducing gain by 10 to 12 dB may result in insufficient output for someone with a severe degree of hearing loss. Figure 1 shows the amplified output of a hearing aid after a 12 dB gain (noise) reduction for someone with a 40 dB hearing loss (left) and a 70 dB hearing loss (right). The black dotted line represents the person’s thresholds. The red line is the amplified noise spectrum, and the green line is the average amplified speech spectrum. The shaded yellow area represents the amplified speech that is above the person’s thresholds and noise spectrum (ie, the audible area). It is clear that the person with the 70 dB hearing loss would have less audible amplified sound. The hearing aid may sound “too soft,” “not loud enough,” or “muffled.” Speech intelligibility may be compromised. The key point is that the wearer’s hearing loss should be included in the design of a noise reduction algorithm.
|FIGURE 1 a-b. Audible area after amplification with a 12-dB gain reduction. The left display (1a) is someone with a flat 40 dB hearing loss, and the right (1b) is someone with a 70 dB hearing loss.|
Preserving Speech Intelligibility: Directional Microphones
1) A low compression threshold (CT). The CT reflects the lowest input level at which a compression hearing aid starts to reduce gain with increasing input levels.9 Assuming that two compression hearing aids have the same gain at a conversational level, the one with a lower CT will typically yield a higher gain for softer sounds. Lee et al10 demonstrated that a fixed directional hearing aid with a low CT yielded higher speech recognition scores in quiet than another hearing aid set to linear and WDRC modes (with a higher CT) for the same conversational gain when speech was presented from the back.
2) Fully automatic. One may provide both an omnidirectional microphone and a directional microphone on the same hearing aid so the wearer can choose the microphone that is the most optimal for the specific situation. However, this may be impractical because the wearers need to know how and when to switch the microphones.
3) Activation thresholds. A fully adaptive directional microphone switches from an omnidirectional mode into a directional mode when its activation criterion is reached. Having a low activation threshold means the hearing aid is directional, even at a low input level. This may be undesirable because the reduced sensitivity of the microphone in the directional mode could further lower the audibility of soft sounds. Kuk et al7 showed the impact of a fixed directional microphone on the audibility for soft speech. To minimize this audibility limitation, the activation threshold for an automatic directional microphone must not be too low. Typically, a level of 50 to 55 dB SPL is optimal. However, the audibility of higher level sounds from the back may still be compromised.
|FIGURE 2. Output spectra over time from the Inteo hearing aid programmed for three degrees of hearing losses (30, 50, and 80 dB HL) when the classic noise reduction algorithm (top row) and Speech Enhancer algorithm (bottom row) are used.|
Additional Contributions from New Processing Strategies
Recently, Widex introduced a new signal processing platform called Integrated Signal Processing (ISP) that integrates the processing of all its features so that the results are shared and utilized by other processing features (see Kuk11 for details). The following discussion focuses on how ISP enhances the design of the noise reduction and adaptive directional microphone systems.
Speech Enhancer (SE) noise reduction. Inteo uses a noise reduction system called the Speech Enhancer (SE) to ensure that its wearers, regardless of the degree of hearing loss, are provided the maximum audibility cues for speech understanding in any noisy environment.12 The SE also includes information on the wearer’s thresholds, the noise level in each of the 15 channels, and the level-appropriate speech spectrum in its gain considerations. It automatically adjusts the available gain settings in each of the 15 channels (over 15 quadrillion combinations of possible settings) and compares each setting combination to the others in order to find one that yields the highest speech intelligibility index (SII)13 without discomfort. In principle, this ensures the best potential for speech understanding.
Because the SE considers the hearing loss of the wearer, different amounts of gain reduction will result for different wearers in the same listening condition. Indeed, those with more hearing loss will probably experience less gain reduction than those with a milder loss. This preserves audibility and further ensures consistent satisfaction from more wearers.
Figure 2 summarizes the output spectra over time measured with a classic noise reduction algorithm (top row) and the SE (bottom row) to a speech-shaped noise. The output for each of the three degrees of hearing losses (30, 50, and 80 dB HL) is shown. For ease of comparison, the level scale is adjusted for different hearing losses. The output is color-coded so that higher output levels are represented by warmer colors, graduating from red (highest output), to yellow, green, light blue, and dark blue (lowest output). For the classic NR algorithm, one can see that all three displays look remarkably similar. Initially, there is a high output that gradually decreases by about 12 dB over time (after 15 s and 60 s). The same decrease is seen for all three degrees of hearing loss.
The output of the SE for a mild hearing loss is less than that from the classic NR; it reduces more gain more quickly. As the degree of hearing loss increases, there is a higher output with the SE than with the classic NR (more red or orange). This may prevent a loss of usable audibility cues in more moderately severe losses.
The SE may automatically increase its gain in noise (the majority of NR algorithms only reduce gain). As much as 5 dB of gain increase is possible if such an increase may further improve the estimated SII without risking feedback and loudness discomfort.
High Definition Locator. The directional microphone system used in the Inteo is a 15-channel fully adaptive directional microphone system. This means that each of the 15 channels can have its own adaptive polar patterns that change from an omnidirectional to various directional patterns. This is different from some multichannel directional systems where the low frequency channel is fixed to an omnidirectional mode.
Why Multichannel Directionality?
Many directional microphones have broadband directivity. This means that the polar pattern is the same for all frequencies. This is similar to a single-channel compression hearing aid where all frequencies share the same gain change even though only one frequency exceeds the compression threshold. The advantage of multichannel compression is to allow flexibility in gain adjustment so that gain reduction may be confined to restricted frequencies. This preserves the audibility of other sounds.
A multichannel adaptive directional microphone provides specificity in the polar patterns. For example, one may have an omnidirectional polar pattern for the 500 Hz channel and a hypercardioid pattern for the 1,000 Hz channel. Furthermore, these polar patterns may change adaptively depending on the acoustic conditions.
One advantage of this flexibility is to have a “quieter” directional hearing aid. This is because a directional microphone that is equalized in its low frequency sensitivity has a higher noise floor than an omnidirectional microphone (see Kuk et al6). This becomes noticeable in quiet. In a multichannel directional hearing aid, the designer can limit the low frequency polar pattern to be omnidirectional and still adaptively vary the polar pattern in the high frequencies. This eliminates the need to compensate for the microphone sensitivity in the lows, thus allowing for a quieter hearing aid in the “directional” mode. The limitation is that it does not offer any directional benefit in the low frequencies, and limits the SNR benefits of the directional microphone.
|FIGURE 3a-b. Waveforms (3a, left) and output spectra (3b, right) from the multichannel fully adaptive Inteo hearing aid (in the omnidirectional mode and adaptive directional mode) and the broadband fully adaptive Diva hearing aid to speech presented in the front and a narrow band noise at 800 Hz presented at 135°.|
Figure 3a shows the output waveform from an Inteo hearing aid in an omnidirectional mode when speech is presented from the front and a 1/3 octave narrow band noise (NBN) at 800 Hz is presented at 135° (left of 3a). This is followed by the output of the Diva broadband directional hearing aid when the same stimuli are presented (middle of 3a). Note the significant reduction in the output waveform. The third waveform (right of 3a) is the output from the Inteo in its fully adaptive directional mode, where the overall magnitude of the output waveform is preserved.
Figure 3b shows the output spectra of the three waveforms from Figure 3a. The green curve shows the output spectrum from the Inteo omnidirectional mic. The purple curve shows the output from the Diva broadband directional mic. Note that the overall output spectrum is reduced by 15 to 20 dB for frequencies as high as 4,000 Hz. As discussed, this is because a broadband directional microphone has the same directional characteristics for all frequencies. The resulting polar pattern (hypercardioid) that is formed to nullify the 800 Hz NBN is also applied to the other frequencies. Speech will sound softer or muffled. Some wearers may report “drowning” of the speech along with the noise. The blue line shows the output from the Inteo IN-9 hearing aid in the adaptive directional mode. Notice that the output at all frequencies but 800 Hz is preserved. This is because only the directivity of the 800 Hz channel is adaptively changed to be less sensitive to the NBN. The directivity at the other frequency channels remains in an omnidirectional mode. The audibility/intelligibility of the speech input is preserved.
ISP further enhances the action of the adaptive microphone by integrating the results of acoustic scene analysis (also used in the noise reduction algorithms) in its switching. When “speech alone” is detected, the hearing aid remains in an omnidirectional microphone mode regardless of its input intensity or its azimuth of presentation. Figure 4 shows the output of an adaptive hearing aid without such a feature (Diva, middle) and the Inteo with such a feature (left in omnidirectional mode and right in adaptive directional mode) when speech is presented from the back (both aids are set to the same settings). The output from the Diva is significantly lower than that of the Inteo because of the lack of speech preservation that is available only in the ISP hearing aids. Indeed, the output from the Inteo in the directional mode is similar to its output in the omnidirectional mode when there is only speech in the environment. This action further preserves audibility in a directional microphone mode.
|FIGURE 4. Output waveform of the Inteo hearing aid (in both omnidirectional mode and adaptive directional modes) and the Diva hearing aid (in adaptive directional mode) to speech presented from 135°.|
Speech Preservation versus Compromise SNR
While the audibility/intelligibility considerations are important, it is important to ensure that the SE algorithm and the multi-channel adaptive directional microphone improve the wearer’s listening in noise. We reported in a previous article on the effectiveness of these two algorithms in a micro-size ISP hearing aid.14 The following reports on its effectiveness in subjects with a moderate-to-severe degree of hearing loss.
|FIGURE 5. SNR improvement on the HINT between 3 and 4 noise sources in the omni+SE and dir+SE condition.|
The speech-in-noise performance of seven adult subjects with a moderate-to-severe flat hearing loss (around 70 dB HL on average) was evaluated. All subjects were experienced hearing aid wearers (3 to 43 years of using hearing aids) with an average age of 66 years (range: 46 to 86 years). Their performance in noise was evaluated with the Hearing in Noise Test (HINT). Continuous HINT speech-shaped noise at 75 dB SPL was presented in two conditions: from the sides and back (90°, 180°, 270°), and from the front, sides, and back (0°, 90°, 180°, 270°). Binaural Widex Inteo IN-19 were used for the testing in an 1) omnidirectional only mode (Omni); 2) omnidirectional plus SE mode (omni+SE), and 3) adaptive directional with SE mode (dir+SE).
Figure 5 shows the SNR improvement for the Inteo omnidirectional microphone (omni) condition. When the noise originated from the sides and back (ie, 3 noise sources), the SNR improvement from the SE was about 1.5 dB, and the combined SNR improvement from the directional mic and the SE (dir+ SE) was over 7 dB (vs the omni mic). When the noise was also presented from the front (ie, 4 noise sources), the SNR improvement was still 2.3 dB with the omnidirectional microphone and 4 dB in the “dir+SE” mode. This suggests that the SE was effective in improving the SNR, and the combination of the adaptive directional microphone and the SE is effective in improving the SNR, although the magnitude of the improvement decreased somewhat when noise originated from the front.
What does this imply? First, in contrast to most studies, which did not report any improvement in SNR by the noise reduction algorithms, this study showed a 2 dB SNR improvement by the SE. This could be the result of the subject population we used in this study; however, similar observations were noted on subjects with a milder degree of hearing loss in a previous study.14 Second, the directional microphone with the SE provided 7 dB of SNR improvement when noise was from the sides and back, and 4 dB even when noise was presented from all directions, including the front. The multichannel adaptivity of the polar patterns may be credited for seeking the best polar response and the SE the best gain settings in such a condition. The use of ISP preserves speech intelligibility and enhances the effectiveness of noise management algorithms.
- Ricketts T, Hornsby B. Sound quality measures for speech in noise through a commercial hearing aid implementing digital noise reduction. J Am Acad Audiol. 2005;16(5):270-277.
- Kuk F, Keenan D, Sonne M, Ludvigsen C. Efficacy of an open fitting hearing aid. Hearing Review. 2005;12(2):26-32.
- Kuk F, Keenan D, Ludvigsen C. Is real-world directional benefit predictable? Hearing Review. 2004;11(11)18-25.
- Valente M, Mispagel K. Performance of an automatic adaptive dual-microphone ITC digital hearing aid. Hearing Review. 2004;11(2):42-46,71.
- Kuk F, Ludvigsen C, Paludan-Muller C. Improving hearing aid performance in noise: challenges and strategies. Hear Jour. 2002;55(4):34-46.
- Kuk F, Baekgaard L, Ludvigsen C. Design considerations in directional microphones. Hearing Review. 2000;7(9):68-73.
- Kuk F, Keenan D, Lau C, Ludvigsen C. Performance of a fully adaptive directional microphone to signals presented from various azimuths. J Am Acad Audiol. 2005;16(6):335-349.
- Marcoux A, Yathiraj A, Cote I, Logan J. The effect of a hearing aid noise reduction algorithm on the acquisition of novel speech contrasts. Int J Audiol. 2006;45(12):707-714.
- Kuk F. Optimizing compression: the advantages of a low compression threshold. In: Kochkin S, Strom KE, eds. High Performance Hearing Solutions, Vol 3: Marketing and Technology. Hearing Review. 1999;[suppl]4(11):44-47.
- Lee L, Lau C, Sullivan D. The advantage of a low compression threshold in directional microphones. Hearing Review. 1998;5(8):30-32.
- Kuk F. Integrated Signal Processing–A new standard in hearing aid processing. Hearing Review. 2006;[suppl]13.
- Kuk F, Paludan-Muller C. Noise management algorithm may improve speech intelligibility in noise. Hear Jour. 2006;.59(4):62-65.
- American National Standards Institute. Methods for the calculation of the speech intelligibility index. New York: ANSI;1993.
- Kuk F, Keenan D, Baekgaard L. Speech in noise performance of a micro-size BTE. Hearing Review. 2007;14(10):60-64.
This article was submitted to HR by Francis Kuk, PhD, director of audiology, and Heidi Peeters, MA, a research audiologist, at the Widex Office of Research and Clinical Amplification (ORCA) in Lisle, Ill. Correspondence can be addressed to Francis Kuk, Widex ORCA, 2300 Cabot Dr, Suite 415, Lisle, IL 60532; e-mail: .