What is a hi-fi hearing aid? Here is a perspective on the considerations that affect the faithfulness in which a hearing aid reproduces original sounds while compensating for the hearing loss of its wearer.
This article was submitted to HR by Francis Kuk, PhD, director of audiology at Widex ORCA, in Lisle, Ill, and Anders Jessen, BS, and Lars Baekgaard, MS, research engineers at Widex A/S in Vaerloese, Denmark. Correspondence can be addressed to HR at or Francis Kuk, PhD, at . |
High-fidelity, or “hi-fi,” typically describes loudspeaker systems that faithfully reproduce the original sounds. Such systems must have a low noise level, are virtually free of distortion, and do not alter the frequency content of the original signal. These criteria are intended to preserve the nuances of the original sounds.
Using such criteria, one would not expect hearing aids to be hi-fi since they purposely shape the spectral input to compensate for the hearing loss of their wearers. Furthermore, they “distort” the sound input to achieve objectives, such as an enhanced signal-to-noise ratio (SNR).
Despite such definitional differences, the extent to which some of today’s digital hearing aids preserve the nuances of the input sounds qualifies them to be described as hi-fi. What follows is a summary of the considerations that would affect the faithfulness in which a hearing aid reproduces the original sounds while compensating for the hearing loss of the wearers. These considerations are included in the design of the new Widex Mind hearing aid.
Retaining Original Sounds and Keeping the Input “Clean”
Sounds differ by their frequency (spectrum), loudness (intensity), and spectral-intensity fluctuation over time. In order to keep the original characteristics of the incoming sounds, a hearing aid must make sure that it preserves as much of the nuances of the input signals as possible.
In a digital hearing aid, the input stage includes the microphone (along with its pre-amplifier) and the analog-to-digital converter (ADC). Each of these components affects the integrity of the input signal.
FIGURE 1. Difference in output when a 1000 Hz tone was sampled at two rates. The top panel shows the case when a sampling frequency of 1333 Hz was used while the bottom panel shows a sampling frequency of 2000 Hz. |
Microphone. A microphone picks up the acoustic signals and transduces them into an electrical form before they are converted by the ADC. Of relevance here is its bandwidth, the noise floor, its directional sensitivity, and its saturation level. The physical bandwidth of today’s microphones is typically broad. This suggests that today’s microphones have the potential of picking up every sound in the environment. However, this may not be as desirable as one may imagine. The very low frequencies—although inaudible to the wearer—may saturate the microphone easily and result in audible distortion. The ultra-high-frequency carriers used in many security systems today may also be picked up by the microphone, resulting in audible artifacts. Consequently, special filters are placed after the microphones to remove the extreme low and high frequencies in order to minimize real-world artifacts.
A microphone is also judged by its polar pattern—its sensitivity to sounds from different azimuths. For a microphone to preserve the nuances of the incoming signals, it has to be sensitive to sounds from all directions. That is, it needs to be omnidirectional. However, a problem with an omnidirectional microphone is that both the desirable and the undesirable sounds are picked up equally. Thus, the SNR encountered by an omnidirectional microphone in noise may be poor. This may make communication difficult for the hearing-impaired wearers. To circumvent the SNR problem, some microphones are designed to have differential sensitivity to sounds from different azimuths (ie, directional). Typically, directional microphones have their maximum sensitivity to sounds from the front, but reduced sensitivity to sounds from the sides and/or the back where more undesirable signals originate.
While a directional microphone may improve the SNR in noise, it achieves its purpose at the expense of a loss in sensitivity to sounds from other azimuths and in the lower frequencies. It reduces the “fidelity” of the natural sounds and may affect audibility and intelligibility of the speech sounds. A solution to such a problem is the use of automatic directional microphones. These are microphones that automatically switch between an omnidirectional mode and a directional mode. This could ensure the fidelity of the sound pick-up in quiet and a higher SNR in noise. Kuk and Peeters1 discussed the various measures to ensure audibility in directional microphones and noise-reduction algorithms.
Microphones saturate when the input level exceeds a particular SPL. Saturation results in distortion products that are included as input to the processor. Such distortions usually sound “crackling or raspy” to the wearers. For typical microphones, this theoretical limit is around 115 dB SPL2 and should be acceptable when conversational or even loud speech is encountered. In practice, the saturation limit is far lower because of the design trade-offs between the noise floor of the microphone and the current consumption required for a high saturation limit. In situations where the microphone has a low saturation limit, the wearer will notice the unpleasant distortion when the input sounds are loud, when vocalizing at a high level, or when playing a musical instrument that produces a high output.
FIGURE 2. Waveforms comparing a linearly amplified output (left) and an amplified output with 2:1 compression (right). |
Analog-to-digital conversion (ADC). In a digital hearing aid, input sounds that are transduced by the microphone go through an ADC phase when the input sounds are converted into a digital (“0” and “1”) form. This conversion process can lead to distortion.
Sampling frequency. The number of times per second the ADC samples the signals and converts them into binary form is the sampling frequency. When the sampling frequency is too low, an insufficient number of the input will be sampled and the true frequency content of the input will be in error. This is called an aliasing error. Figure 1 shows a 1000 Hz input sampled at two rates. In the top case, a sampling frequency of 1333 Hz was used. This resulted in a reconstructed signal of only 333 Hz. This results in an altered sound quality. In the bottom case, a sampling frequency of 2000 Hz was used. This results in an accurate representation of the input frequency.
To ensure adequate sampling, the Nyquist theorem states that one must sample, at minimum, twice the highest frequency in the sample. For example, if one knows that a signal includes energy up to 8000 Hz, one should sample at 16,000 Hz to make sure that all the frequencies in the input are sampled. For hi-fi processing, it is often necessary to sample at a higher frequency than 16 kHz because a significant amount of speech information is available above 8000 Hz. For example, Boothroyd and Medwetsky3 reported that the phoneme /s/ produced by female speakers has significant energy above 8000 Hz.
One should also remember that, unless the input is sampled at a high enough sampling frequency, it is meaningless to have a receiver that claims to have an extended bandwidth. For example, a sampling frequency of 16,000 Hz does not permit input beyond 8000 Hz to be processed correctly by the hearing aid. Thus, even when the receiver has an extended bandwidth up to 10,000 Hz, the effective output from the hearing aid would be limited to 8000 Hz.
Quantization. The sampled signal needs to be converted into a form that can be processed. The mapping of the intensity information is called quantization. The accuracy of the intensity mapping is highly dependent on the design of the converter and its bit resolution. In general, the more bits there are, the higher its resolution. As an approximation, each bit provides about 6 dB resolution. Thus, for a typical ADC that uses 16 bits, it will have a resolution of 96 dB. Sounds exceeding 96 dB SPL could be severely distorted unless one designs to accept a poorer resolution for the softer sounds.
To avoid saturation at the input stage from common everyday loud sounds, one can do one of two things (other than changing the ADC design).
- Increase the number of bits of the ADC. This is achieved at the expense of increasing the current drain and power consumption.
- Place a compression limiter at the input stage. This is a “protector” circuit that typically has a high compression threshold and a high compression ratio (typically 10:1 or higher). This circuit will ensure that the input to the ADC does not exceed the saturation limit. Doing so will reduce the occurrence of saturation distortion without incurring too much additional power consumption. This is the approach taken by many manufacturers to minimize saturation distortion at the input stage.
There are some drawbacks to this approach. A compression limiter typically limits the input to the ADC at a low, fixed value even though its level increases. For example, let’s assume that the CT is set at 90 dB SPL and an input of 100 dB SPL is received at the microphone opening. The input going into the ADC will be 91 dB SPL; and an input of 110 dB SPL will appear as 92 dB SPL. This means that the intensity changes (or fluctuations) at the high-input levels are reduced substantially! This intensity smearing masks potential intensity cues so they will not be easily available for processing. The reduction in the naturalness of the input may be one reason why speech understanding in real-world, loud, noisy situations is difficult for hearing aid wearers. Obviously, the matter may be even worse if no means of limiting the input, or peak clipping, are in place.
A second drawback is that the level of the maximum available input varies from a typical level of around 90 dB SPL to as high as 107 dB SPL in some devices. This suggests a wide range of performance differences among manufacturers. Chasin4 suggested a level of 105 dB SPL to be acceptable for music appreciation. While this information may be available from the manufacturer of the specific hearing aid, one may also estimate this level oneself. To do this, Chasin4 suggested programming the hearing aid to a minimal gain but maximum output setting. All the adaptive features (such as noise reduction, directional microphone, etc) should be deactivated to minimize any biases. A sinusoid of increasing intensity level should be presented to the hearing aid. The input level at which a significant increase in distortion (when peak clipping is used as the input limiting mechanism) or no increase in output (when a compression limiter is used as the limiting mechanism) occurs should be noted. This is the maximum input level.
The advantage of a high input saturation level is the preservation of the natural input (including speech) in its entirety for accurate processing. The risk for saturation distortion and intensity/temporal distortion increases with a low input saturation level. This results in poor sound quality with user complaints such as “crackling” or “raspy” at loud inputs. Unfortunately, no amount of gain adjustment on the hearing aid can correct for the problem because this is an input stage problem. Musicians who play their instruments at a high level and wearers who vocalize at a loud level may be especially affected.
FIGURE 3. Hypothetical difference in output waveforms between fast-acting and slow-acting WDRC. A) Input-output characteristics of the WDRC circuit; B) Hypothetical input (intensity over time); C) Output when short attack and release times are used; D) Output when long attack and release times are used. |
Keeping the Processing as Linear as Possible
Hearing aids perform constructive distortion. That is, they alter the input sounds so they can be audible and intelligible to the wearers in as many environments as possible. Frequency shaping compensates for the differential loss of hearing across frequencies. Compression compensates for the change in loudness growth function. Noise reduction reduces the amplified output to maintain comfort. These algorithms are necessary even though the input signals may be distorted. However, if any of these processing steps can be avoided without a loss of audibility and intelligibility, they should be considered. Thus, algorithms like noise reduction and directional microphones should not be active in quiet in order to preserve the naturalness of the signals.
Slow-acting WDRC. One processing algorithm that is active at all input levels is compression. Thus, the operation of this algorithm determines how it may preserve the naturalness of the input. As reviewed in Kuk,5,6 linear processing preserves the intensity relationship of sounds (ie, its naturalness). Linear hearing aids, while having the advantage of linear processing, have the limitation of insufficient audibility for soft sounds and discomfort for loud sounds without manual volume control adjustment by the wearer. In contrast, wide dynamic range compression (WDRC) hearing aids provide audibility for the softer sounds and comfort for the louder sounds without wearer VC adjustment. But the action of compression reduces the intensity relationship of sounds, especially for multichannel hearing aids. Figure 2 shows a comparison in output between a linear hearing aid and a compression hearing aid that has a 2:1 compression ratio and an attack and release time of 1 ms. The intensity variation seen in the linear hearing aid is clearly absent in the WDRC hearing aid.
The reduction in the intensity relationship in the previous example could result in an “unnatural,” “muffled,” or “noisy” sound perception. But WDRC does not always lead to a reduction in the intensity contrast of the input signals. It does so only in hearing aids that use moderately fast attack and release times (or fast-acting). This includes systems that use attack times below 10 ms and release times below 50 ms.5 These time constants are intended to allow the hearing aid gain follow the short-term variation in speech intensity during conversations.
Figure 3a shows the hypothetical static input-output curve of a WDRC hearing aid with a CR of 2:1 and a CT at 50 dBSPL. Figure 3b shows a hypothetical input with two intensity peaks at 80 dBSPL and 60 dBSPL (or a peak-to-peak difference of 20 dB). Figure 3c shows the output of the WDRC hearing aid when it uses a fast attack and release time. In this case, the noise floor would be amplified to 70 dBSPL and the two peaks would be amplified to 95 dBSPL and 85 dBSPL. This would reduce the inter-peak difference to 10 dB. Furthermore, the difference between the louder sound and the noise floor is reduced from 40 dB (80 dB – 40 dB) to 25 dB (95 dB – 70 dB). The intensity variation seen in the input signal is not preserved.
A long attack and release time (or slow-acting compression, SAC), which typically means that the attack time is longer than 20 ms and the release time is longer than 1 second, may better preserve the intensity relationship over time. Figure 3d shows the hypothetical output from the same input (using the same input-output curve) with a much longer attack/release time. In this case, the peak-to-peak difference between the more intense and the less intense signals is noted at 19 dB, and the peak-to-floor difference is in excess of 35 dB. Thus, the intensity relationship of sounds is better preserved. Indeed, this is the advantage of a slow-acting WDRC—better preservation of intensity envelope while achieving level adjustment over time. It acts like a linear hearing aid in the short-term because it preserves the instantaneous intensity differences. It acts like a WDRC hearing aid in the long-term because it automatically adjusts its gain over time without any manual VC adjustment. In order words, a SA-WDRC has the benefits of both linear and WDRC hearing aids.
The subjective impression is improved when the intensity relationship of the input is maintained. For example, Neuman et al7 showed that hearing-impaired people preferred a hearing aid with a longer release time. Hansen8 reported greater preference for a longer release time in a multichannel WDRC hearing aid. Souza and Kitch9 also indicated that the amplitude envelope plays an important role in speech understanding.
A potential drawback of using slow-acting WDRC is the sub-audibility of soft sounds that may occur after a loud sound. This is because the gain assigned to the loud sound may not recover fast enough to be sufficient for the soft sounds. Consequently, a quick readjustment of gain will be necessary to ensure audibility. The use of an adaptive compression system, one that adaptively changes its attack and release times, will be desirable to overcome this limitation. A slow-acting WDRC with adaptive release times would preserve the natural intensity fluctuation of the input signals while minimizing the risk of sub-audibility.
FIGURE 4. Actions of a typical compression limiter (CL). The top panel shows the steps of a compression limiter. The reaction time is the time it takes the CL to achieve gain reduction. The bottom panel shows the output waveform even with a compression limiter. The areas in red indicate the momentary peak clipping. |
Keeping the Output Full and Clean
The bandwidth of the receiver and the output limiting method used by the hearing aid affect the fidelity of the output sounds.
Receiver bandwidth. Hearing aids are miniature amplifier-loudspeaker systems. The bandwidth of the receiver/loudspeaker limits the range of sounds that is delivered by the hearing aid. Typically, a BTE hearing aid receiver has a bandwidth that extends to 6000 Hz, while a custom ITE may extend to 8000 Hz. Recently, hearing aid receivers that extend beyond 10,000 Hz have become available.
It is reasonable to expect that a broader bandwidth results in a fuller spectrum of sounds at the output. This could improve the “naturalness” and sound quality of the amplified speech. Other benefits of a fuller spectrum include improved localization in the vertical plane10 and improved perception and production of high-frequency speech sounds. For example, Stelmachowicz et al11 stressed the importance of high-frequency audibility in the identification and production of fricative sounds in children.
The advantage of an extended bandwidth in adults may not be universal. The studies so far indicated that, while normal-hearing adults preferred an extended bandwidth, hearing-impaired listeners showed little consistent preference.12 Those with severe losses seem to report less benefit than those with a milder loss. In some cases where the hearing loss in the high frequencies is severe, increasing bandwidth actually decreased speech understanding because of the presence of “dead” regions.13 This suggests that, while a broader bandwidth may be appreciated by normal-hearing people or those with a mild hearing loss, its preference may be variable in people with more than a moderate hearing loss.
Output limiting. Saturation distortion also occurs when the output of the hearing aid exceeds the saturation limit of the receiver (or output A/D converter). The hearing aid would sound “raspy,” “crackling,” or “noisy” to the wearers when this occurs. Although most hearing aids today have a good way of managing high output, it is estimated that the majority of hearing aids sold even 15 years ago had peak clipping as their primary mode of output limiting.
Most modern digital hearing aids use a compression limiter at the output stage to prevent saturation distortion. Typical of a compression limiter, the CT and the CR of the output compression limiter are high. In order to be responsive to the loud sounds that may saturate the receiver, the typical compression limiter uses short attack and release times. To be cost effective, a broadband limiter is used. This means that one frequency within the output that exceeds the compression threshold of the limiter would lower the gain (and output) for all frequencies.
There are two potential shortcomings with the typical compression limiter. First, the use of a broadband limiter lowers the output of all frequencies even though only a small frequency region exceeds the saturation limit. This reduces the audibility of all sounds and can make understanding and the perception of loudness at high input levels more difficult. To overcome this problem, narrow-band limiting is desirable so each channel is responsible for its own output regulation. This ensures audibility and intelligibility in some noisy situations.
The other shortcoming is the activation time of the compression limiter. As indicated, almost all compression limiters are fast-acting systems so they can be responsive to the fluctuations in the output. Unfortunately, even when the attack times may be brief, distortion can still occur before the limiter is in full compression. Figure 4 demonstrates this situation. The upper panel shows the three steps that we have discussed. In Step A, the output of the hearing aid exceeds the pre-set clipping limits. In Step B, the onset of the signal triggers the compression limiter to go into gain reduction. The time it takes for the limiter to reduce gain is indicated by the “reaction” time of the device. In Step C, the portion of the signal during the “reaction time” is not reduced in amplitude. Thus, that portion of the signal would create saturation distortion. The bottom panel in Figure 4 shows the waveform over time with the conventional compression limiter. The dark areas indicate that the compression limiter is active, while the areas in red represent the momentary saturation distortion because the limiter is not in full gain reduction.
FIGURE 5. Output waveform of speech in hammering noise background with (bottom panel) and without (top panel) TruSound Stabilizer. |
Processing with Hi-Fi in Mind
The considerations that are important to yield hi-fi sounds are included in the TruSound feature of the Mind440 hearing aids. The special collection of features ensures that the amplified sounds are as true a replica of the original input as possible (while achieving audibility). This requires special designs at the input stage, the signal processing stage, as well as the output stage.
At the input stage, the Mind440 hearing aid uses two omnidirectional microphones to form a fully adaptive multichannel directional microphone. The microphone keeps its omnidirectional polar pattern in quiet and when speech is the only input. The processor uses a 32,000 Hz sampling frequency to capture up to 16,000 Hz of the input signal. A sigma-delta converter is used to convert the analog signal into its digital form. No compression limiter is used, but linearity of the input is maintained up to 107 dBSPL to avoid smearing of temporal fine structures.
The Mind440 uses primarily SAC to preserve the intensity contrasts and the temporal fine structures within the input signals. To overcome the potential of sub-audibility, a TruSound Stabilizer is available, which adaptively changes the attack/release times of the hearing aid based on the duration and intensity of the input signal. In typical situations where the intensity fluctuation is not large, SAC is used to retain the normal intensity relationship. In situations where impulse sounds occur, TruSound Stabilizer shortens its attack and release times to preserve audibility of the softer sound and achieve comfort of the louder, impulse sounds. Figure 5 shows a comparison of the waveforms of speech embedded in an impulse sound background (hammering). The speech signal is more visible (intelligible) with the TruSound Stabilizer.
Mind440 has made a special effort to ensure the output is as broad and as distortion free as possible. For example, the micro-version of the hearing aid uses a two-way receiver that provides a frequency response that ranges from 100 Hz to 10,000 Hz (in situ). This ensures that wearers who can benefit from the extended bandwidth enjoy the extra high frequency amplification.
Another unique feature of the device is the use of TruSound AOC (automatic output compression or compression limiting). This output limiting system has two components. A narrow band system automatically controls the output from each of the 15 channels independently. A broadband system manages any unforeseen or unanticipated output; it acts as the final output control. The system manages compression limiting (or gain reduction) differently than traditional compression limiters. Rather than responding to the peak output when it occurs, TruSound AOC uses a pre-peak detector that measures the occurrence of the peak before the peak output occurs. This allows the compression limiter to start reducing gain before the actual gain reduction is needed.
Figure 6 shows the sequence of action for this pre-peak detection, and the associated waveform that is processed with TruSound AOC. Compared to the previous situation where slight distortion occurs (Figure 4, page 41), pre-peak detection virtually yields no output distortion.
In summary, high-fidelity processing in hearing aids requires careful considerations of each step of the signal pathway and using signal processing algorithms to preserve the natural spectral-temporal content of the input. The Mind440 TruSound algorithm is designed to exemplify these considerations.
FIGURE 6. Pre-peak detection in TruSound AOC. The upper panel shows that gain reduction is activated prior to the onset of the peak output. The lower panel shows no trace of any saturation distortion in the output waveform. |
References
- Kuk F, Peeters H. Speech preservation in noise management strategies. Hearing Review. 2007;14(12):28-40.
- Chasin M, Russo F. Hearing aids and music. Trends Amplif. 2004;8(2):35-47.
- Boothroyd A, Medwetsky L. Spectral distribution of /s/ and the frequency response of hearing aids. Ear Hear. 1992;13(3):150-157.
- Chasin M. Can your hearing aids handle loud music? A quick test will tell you. Hear Jour. 2006;59(12):22-26.
- Kuk F. Theoretical and practical considerations in compression hearing aids. Trends Amplif. 1996;1(1):1-39.
- Kuk F. Rationale and requirements for a slow acting compression hearing aid. Hear Jour. 1998;51(6):41-53, 79.
- Neuman A, Bakke M, Mackersie C, Hellman S, Levitt H. The effect of compression ratio and release time on the categorical rating of sound quality. J Acoust Soc Amer. 1998;103(5):2273-2281.
- Hansen M. Effects of multi-channel compression time constants on subjectively perceived sound quality and speech intelligibility. Ear Hear. 2002;23(4):369-380.
- Souza P, Kitch V. The contribution of amplitude envelope cues to sentence identification in young and aged listeners. Ear Hear. 2001;22(2):112-119.
- Byrne D, Sinclair S, Noble W. Open earmold fittings for improving aided auditory localization for sensorineural hearing losses with good high-frequency hearing. Ear Hear. 1998;19(1):62-71.
- Stelmachowicz P, Lewis D, Choi S, Hoover B. Effect of stimulus bandwidth on auditory skills in normal-hearing and hearing-impaired children. Ear Hear. 2007;28(4):483-494.
- Hornsby B, Ricketts T. Hearing aid bandwidth for speech understanding. Workshop presented at 2007 Convention of the American Academy of Audiology, Charlotte, NC.
- Moore B. Dead regions in the cochlea: diagnosis, perceptual consequences, and implications for the fitting of hearing aids. Trends Amplif. 2001;5(1):1-34.
Citation for this article:
Kuk F, Jessen A, Baekgaard L. Ensuring high-fidelity in hearing aid sound processing. Hearing Review. 2009; 16(3):34-43.