|
|
Solutions of Apherma Hearing Aids Hearing loss, particularly sensorineural loss, is a complex disorder. Hearing aids should not be expected to fully restore the function of impaired ear. Nevertheless, good hearing aids are expected to provide valuable assistance to patients under most situations. The Apherma hearing aids are designed to offers superior performance for a majority of patients in a wide range of listening environments. With advanced DSP algorithms, the Apherma hearing aids can help the patients:
A. Dynamic range For sensorineural hearing losses, hearing aids need to compensate not only for the reduced sensitivity but also for the reduced dynamic range. As a result, while the patients can hear soft sounds, their loudness sensation of intense sounds would be similar to that of normal listeners. Apherma hearing aids achieve these expectations through a comprehensive solution that compensates the various deficiencies associated with the loss of dynamic range. A1. Multiband Wide Dynamic Range Compression For Hearing Loss Compensation Apherma's multi-band wide dynamic range compression (WDRC) is designed to overcome the lost dynamic range associated with hearing loss. It separates sound into six frequency bands, and adjusts the gain in individual bands according to the patient's residual dynamic range still available in those bands. Within individual bands, this processing mimics the nonlinear compression function of the outer hair cells in a normal cochlear. It provides greater amplification to softer sounds. Thus, for both soft and loud sounds, the loudness that hearing-aid users will perceive will be similar to that perceived by the normal listeners. As a result, the algorithm helps recover the dynamic range of impaired ears to as close to the normal range as possible. An important goal of the Apherma WDRC system is to enhance speech understanding. Equipped with fast time constants, the system can react to rapid changes in the speech envelope, allowing it to provide a sufficient boost of soft consonants while not over amplifying the more intense vowels. As a result of this ‘syllabic compression’, all parts of syllables can be heard in harmony. A2. The Flexibility Of Selectable Time Constants Time constants of a wide dynamic range compression (WDRC) system describe how fast the compressor reacts to changes in the input stimulus level. There is no lack of assumptions on how the time constants should be set, but the conclusions often disagree. Those argue for fast action, or short time constants, are based largely on two reasons. One comes from a physiological point of view that the cochlear nonlinear compression provided by outer hair cells is fast acting. As WDRC is primarily intended to restore the lost outer-hair-cell function, it should be also fast acting. The second reason is from a linguistic point of view. A short time constant allows the sufficient amplification of consonants while not over amplifying vowels, providing the “syllabic compression” suited for speech inputs. There are also practical reasons against slow action. For example, if reaction of the compressor to input level change is too slow, a big gain which is suitable for soft sound could be added to an intense sound, and the result is an overly loud sound that last longer than anyone is willing to tolerate. Those who prefer a slow acting compressor focus on mostly empirical evidence that many patients report a more comfortable listening experience with a relatively slow compressor. Faster time constants are usually accompanied by increased amount of artifacts and distortions in the spectro-temporal structures of signals, particularly that of speech. Obviously, the fast-versus-slow acting compression is a complicated issue subject to further research. Rather than making any pre-assumption and taking a firm stand on the issue, Apherma offers dispensers the freedom to select either fast or slow time constants, whichever allows them to optimize the hearing-aid performance and comfort for individual patients. In addition to the choice between fast and slow actions, Apherma offers the dispenser an option of dual time constants¡ªthe slow time constant for low-frequency channels and the fast time constant for high-frequency channels. This scheme considers the fact that the relatively narrow bands at low frequencies have an inherently slow fluctuation of the signal envelope. Using the slow time constant in those channels avoids artifacts and distortions sometimes associated with fast compression, without losing much information regarding level fluctuations. At high frequencies, on the other hand, the bands are relatively wide and envelopes fluctuate fast. These channels also contain critical speech information. Using the fast time constant in high-frequency channels allows the compressor to follow the envelope fluctuations at the syllable level, performing syllabic compression. By offering these flexibilities to the dispenser, Apherma believes that the patient will have the best chance to walk away with a device that will best serve his or her need. A3. Spectral Re-Mapping (SRM) When the hearing loss in a frequency region is too severe, the small residual dynamic range renders WDRC ineffective. Some of these losses may actually reflect a ‘dead’ cochlear region where neural receptors, the inner hair cells, or the auditory nerves are completely gone. Research evidence suggests that amplification in frequency regions of severe losses can actually be detrimental to speech understanding. Yet, such regions typically occur at high frequencies containing crucial speech information. We must find new solutions to retrieve the important speech information from those regions and make it available to the patient.
The solution by Apherma is a spectral re-mapping (SRM) algorithm. It continuously analyzes input stimulus for phonemes. When significant speech energy falls into the frequency region of severe loss, SRM transposes that part of the energy to a lower frequency region where the hearing is normal or no more than a moderate loss. Although the original information at the lower frequency location is lost by the SRM processing, we must remember that speech information at higher frequency locations are more important for understanding speech. B. Distortion B1. Spectral Fidelity Preservation (SFP) To Prevent Spectrum Distortion Associated With WDRC Within individual frequency bands, WDRC overcomes the impaired dynamic range by compressing the stimulus into the hearing range still available. Across bands, however, WDRC could distort the original spectral shape. Suppose the signal levels in two adjacent bands are 50 and 70 dB respectively, with a spectral contrast of 20 dB. After the WDRC applies a 30 dB gain to the first band and 15 dB to the second, the spectral contrast is reduced to only 5 dB. This effect of smearing spectral contrast is a general problem with WDRC. As spectral contrast often signifies speech sounds, a dramatic change in the contrast could present significant distortion, affecting the quality of speech stimuli. Apherma devices employ an algorithm of Spectral Fidelity Preservation (SFP, patent pending) to overcome this spectral smearing problem. Its function is to optimize audibility for soft speech and preserves phonemic signature for normal and loud speech. For the signal of the above example, the combination of the Apherma WDRC and SFP would produce a contrast of about 15 dB, largely preserving the original spectral contrast. With this unique feature, the Apherma aids produce sounds of unparalleled clarity. B2. Soft Level Limiter for Hearing Aid Output Like all audio devices, hearing aids are optimized within a certain dynamic range, where they operate with minimal distortion. When over driven, hearing aid output will distort. Distortion from overload is a significant reason for user dissatisfaction. To avoid overload, some hearing aids impose a ceiling to limit the signal peak amplitude, called peak clipping. However, clipped sound is itself an annoying distortion. In Apherma devices, the overload is avoided using a ‘softer’ form of level limiter whose distortions are much less objectionable than that produced by peak clippings. C. Feedback Management Although very often patients could benefit from some additional gain in their hearing aids, many practical issues limit the maximum gain that can be achieved. An important factor is the occurrence of instability or oscillation due to the presence of feedback signal. Between the hearing aid microphone and the receiver there exists acoustic path, from either the air vent, or imprecise fit between the ear mold of the aids and the ear canal. Through this path, amplified sound can travel out of the canal and be picked up by the microphone, and be amplified again. When the net gain of this loop exceeds the net attenuation, the system becomes unstable and will oscillate, producing a high-pitched whistle. Few who hear such whistle would tolerate it for long. To avoid oscillation, the gain must be kept below certain value. On the other hand, hearing aid designers constantly look out for ways to add more gain, even for just several decibels. That extra gain can translate into improved audibility and speech understanding. Effective algorithms can create extra gain, or headroom, of more than several decibels.  Fig. 4. Feedback management with Apherma¡¯s effective AEC algorithm. Without AEC, the system becomes unstable when gain boost occurred mid way through the waveform (top panel). With AEC, the system remains stable throughout the waveform (bottom panel). The solution of the Apherma aids is a very effective Adaptive Echo Cancellation (AEC) algorithm. In this algorithm, an adaptive filter is used to generate an anti-feedback signal in an attempt to cancel the feedback while leaving the direct signal intact. As a result, gain can be increased further without causing feedback. This increase in maximum attainable gain, or headroom, varies in different ears, but can exceed 20 dB in some cases. D. Noise D1. Automatic Microphone Array (AMA) To Improve Signal-to-Noise Ratio and Speech Intelligibility 
Fig. 5. Tracking and elimination of dominant noise source by Apherma's AMA directional algorithm. The system tracks down and eliminates the dominant noise source, as indicated by the direction of the null of the polar pattern when the noise is behind the hearing aid user (a) or at 120 degree to the right (b). Ranked high among difficulties associated with hearing loss is communication in the presence of background noise. To understand speech in a noisy environment, the hearing impaired would require higher signal-to-noise ratio than people with normal hearing. The greater the loss, the higher the signal-to-noise ratio is needed. Simple amplification would not be of much use, because the noise is amplified the same way as the speech, leaving their ratio unchanged. What we need is some scheme that can process speech and noise selectively, and improve signal-to-noise ratio, thus speech intelligibility. In general, this is a very difficult task. In the special cases where the signal and noise come from different directions, signal-to-noise ratio can be improved using directional devices. Such devices have a sensitivity that is direction dependent, as illustrated by plots of polar patterns (see Fig.5). By orienting the device's direction of the maximum sensitivity toward the target source (e.g., a talker), unwanted sources at other directions will be attenuated. Directional algorithms are highly valued in the hearing-aid industry, because no other schemes can consistently and reliably improve signal-to-noise ratio like they do.  Fig. 6. Elimination of dominant noise source by Apherma¡¯s AMA directional algorithm. Much of the noise in the original speech-plus-noise waveform (first part of the waveform) is eliminated by the directional algorithm, producing a clean speech waveform (second part of the waveform). Automatic Microphone Array (AMA) is the patent-pending directional algorithm of the Apherma devices. The AMA uses two microphones to enhance signals from the front while attenuating sound from other directions. It can automatically track and suppress the dominant noise source. It continuously monitors and compensates for changes occurring between the two microphones. This is very important, as even a small mismatch in microphone responses can quickly compromise the effectiveness of the algorithm. Finally, when directional listening is not necessary, the algorithm can turn off itself to make the hearing aid receptive to information from all directions. Equipped with this effective, sophisticated, and powerful algorithm, the Apherma devices will provide the patients a truly rewarding listening experience. D2. Multi-band Noise Reduction (MNR) Aside from causing difficulties in speech understanding, background noise also causes listening discomfort and auditory fatigue. Depending on its temporal and frequency relationship with speech signal, noise can be separated into two groups of segments. Those that overlap with speech signal in both time and frequency are called the simultaneous noise segments. The rest are called the non-simultaneous noise segments. Although a reduction of noise in either group will increase the overall signal-to-noise ratio, only reduction of the simultaneous noise will actually improve speech intelligibility. It is mostly the simultaneous noise that does the actual masking of speech and deteriorates speech intelligibility. To deal with this group of noise segments, Apherma devices rely on the automatic-microphone-array (AMA) algorithm, as described in the previous section. On the other hand, the non-simultaneous noise segments contribute little to the masking of the speech, and they mostly cause listening discomfort. The primary goal of the Apherma Multi-band Noise Reduction (MNR) algorithm is to eliminate this group of noise segments, and to improve listening comfort. The MNR algorithm distinguishes speech from noise by analyzing the statistical properties of the signal in individual frequency bands. For a given band, when the analysis indicates a dominance of speech, the signal of that band is let pass intact. When the analysis indicates a dominance of noise, the sound in that band is attenuated. Overall, the noise segments that fall in speech pauses will be reduced significantly, leaving a much cleaner speech. 
Fig. 7. Attenuation of noises by Apherma¡¯s MNR algorithm. The speech waveform becomes much cleaner with the MNR algorithm activated (second half of the waveform) than without it (first half of the waveform). Through extensive research and testing, the Apherma noise-reduction algorithm is now equipped with optimized parameters. By providing a maximum noise reduction while keeping the signal-processing artifact to the minimum, this effective algorithm will truly enhance the listening experience for users of the Apherma devices. D4. Gain Expansion To Reduce Ambient And Circuit Noise Gain expansion is designed to manage noises of relatively low levels. Such noises include ambient noises from outside sources, and circuit and microphone noises generated inside the hearing aids. For people with normal hearing, ambient noises can be easily ignored in most cases. For hearing aid users, on the other hand, without modification to the amplification scheme for very soft sounds, both ambient and hearing-aid noises are amplified, creating a noisy and unpleasant situation even in a relatively quiet environment. To address this problem, Apherma's WDRC includes a modification to the gain function for low input levels. In this gain expansion algorithm, the gain value for very soft sounds is reduced with decreasing input level. As a result, very soft sound (including noises from the hearing aids) will stay soft. Sounds with levels above the expansion knee will be largely unaffected by this operation and amplified in a normal manner. Most speech sounds that are audible to normal listeners should be largely unaffected. The automatic gain expansion algorithm helps improve the comfort level for hearing aid users. |
