Noise Reduction in Advanced Technology Hearing Aids

Donald J. Schum, Ph.D./CCC-A
Vice President, Audiology & Professional Relations

Douglas L. Beck, Au.D.
Director, Professional Relations
Oticon, Inc., Somerset, New Jersey

Introduction:

Digital signal processing allows algorithms to detect and identify the speech and non-speech components of sound. Once identified, advanced technology hearing aids can be adjusted to de-emphasize frequency regions and time periods dominated by noise, resulting in reduced loudness (Schum, 1996) and decreased annoyance (Schum, 2001) from background noise.

Occasionally, the true operation and impact of noise reduction systems have been misunderstood by professionals and patients. Unfortunately, the name 'noise reduction' implies to some, these systems eliminate noise while leaving the speechsignal intact. Perhaps this misunderstanding is based on 'noise cancellation' systems such as those used in modern headphones which can indeed eliminate vast quantities of unwanted noise. However, sound-based signal cleansing of that magnitude (i.e. noise cancellation) is not possible at this time (2006) in hearing aids. Nonetheless, modern noise reduction systems in hearing aids do provide benefits to the patient.

In this paper, design principles of noise reduction systems in modern hearing aids, and reasonable expectations for the patient, will be discussed.

What are noise reduction systems supposed to do?

The purpose of noise reduction systems is to improve the acceptance of hearing aid amplification in noisy environments. Noise reduction systems are almost exclusively implemented in multi-channel, nonlinear (MCNL) products. MCNL processing provides increased gain for soft sounds, while maintaining medium sounds at a comfortable loudness, while reducing gain for louder inputs to assure listening comfort (Dillon, 2001). However, prescriptive formulae for MCNL systems often assume the sound input is speech and the sound output (from the hearing aid) is delivered into a sealed cavity (the cavity between the tympanic membrane and the medial end of the hearing aid). Therefore, prescribed gain is often based on the presumed desired audibility of speech sounds. Unfortunately, the input is often not speech (i.e., noise) or is a mixture of speech and noise. When the input contains non-speech, patients prefer to have the signal levels reduced and therefore, noise reduction systems were developed.

When the background is comprised of noise only, effective noise reduction systems reduce the noise level to make sound more comfortable for the listener. When the environment is comprised of a mixture of speech and noise, the role of noise reduction is to decrease the loudness and annoyance of the noise, without unduly reducing available speech information.

How do noise reduction systems work?

Two basic and independent functions are common across all noise reduction systems: Analysis and Response.

Analysis: The system will evaluate the incoming signal to determine the presence of speech or noise. In most systems, this analysis is performed on a frequency specific basis.

Response: After the analysis has been completed, the circuit directs the noise reduction system to implement frequency-specific changes in gain.

Separating speech from noise:

Unfortunately, simply confirming that speech exists within a speech-in-noise sound source, does not necessarily mean speech and noise can be separated from each other. Importantly, noise reduction systems do not improve the signal-to-noise ratio (S/N) in a given frequency region. Gain reductions to reduce noise sources in a given spectral region will be applied to speech energy in that same frequency region. This is likely a key source of confusion with regard to noise reduction systems.

Modulation Detection:

Modulation detection is a common noise analysis technique. Modulation Detection determines if the on-going envelope of the sound signal looks more like speech or more like noise (Schum, 2003).

As shown in Figure 1, compared to certain, common background noises, naturally occurring speech has periods of higher and lower intensity (upper panel). Native English speakers normally produce between 4 and 8 syllables per second (lower panel) with intensity peaks that correspond to the louder, vowel sounds within the syllables. Weaker, unvoiced consonants and pauses between syllables and words account for intensity valleys. Comparing the envelope of clean speech to the envelope of background noise, to the envelope of speech-in-noise (Figure 2) reveals clear contrasts. Modulation detection determines how much the signal envelope resembles speech, noise, or a combination of speech-in-noise. Modulation detection is usually performed within each channel of modern advanced technology MCNL hearing aids.

Synchrony Detection:

Synchrony detection is less common, but very important (Elberling, 2002) with regard to noise reduction systems and sound signal analysis. Voiced-speech has a dominant, low-frequency periodic component: the fundamental frequency of the voice. However, voiced-speech also contains energy across upper harmonics (multiples of the fundamental frequency) and in that respect, the fundamental frequency is apparent across the entire speech spectrum. Figure 3 shows a segment of voiced-speech filtered into four different high-frequency bands. As can be seen, there is a common, low-frequency component across the high-frequency bands.

Synchrony detection examines the moment-to-moment morphology of the waveform across the high frequency bands to determine if there is a common, dominant low-frequency component. If a common low-frequency component is detected, it is presumed the sound signals across the high-frequency bands share a common, low-frequency source. Since speech and music are the two principle sources of sound based on a harmonic structure (as described above), Synchrony detection and analysis can be used to identify speech in the acoustic environment.

Where are noise reduction systems least effective?

Noise reduction algorithms are criterion-based systems. In other words, incoming signals must meet certain criteria for a response to occur. The ability of these systems to accurately identify speech and noise is dependent on the specific nature of the sounds in the acoustic environment.

Interestingly, because the noise reduction systems described above (synchrony and modulation detection) differentiate between speech and non-speech signals, noise reduction will have little or no effect when the noisy competition is one or two other talkers. The patient will likely experience difficulty in such an environment. Although the patient may consider the other talkers as noise, noise reduction systems cannot identify and separate a nearby target talker from nearby "noisy" talkers. Directional microphone systems can help reduce the background noise in these situations, if the location of the target talker and the competing talkers are spatially separated.

Gain Reduction and Speech:

When a noise reduction system determines noise is present in one or more channels, the quantity of gain reduction is based on several factors. First, gain reduction is typically based on the overall input level. As noted above, the purpose of noise reduction systems is to improve the acceptance of hearing aid amplification in noisy environments. Therefore, it makes sense that gain reductions should be initiated in louder environments. Second, the quantity of gain reduction should, and can be, be varied across frequency.

Given a speech-in-noise acoustic environment, gain reductions predicated on the presence of noise within a given frequency band will reduce speech information within that band. However, to prevent the loss of speech information, while attenuating noise, speech-oriented frequency regions can be protected, with little or no gain reduction applied. In certain advanced technology hearing aid applications, noise reduction algorithms exist which minimize gain reductions across frequency regions in which critical speech information resides. The critical speech sound regions have been determined based on the Articulation Index (AI) (Kryter, 1962) and are well known to audiologists, speech scientists and acousticians. The most critical speech sound regions are from 1000 through 3000 Hz, consistent with the second formant of speech (F2).

Synchrony detection algorithms provide a binary output decision based only on the presence or absence of speech. Gain reductions are based on this straightforward analysis. In the first commercial application (Flynn, 2003), synchrony detection aggressively applied gain reduction in the absence of speech. However, when speech was present, gain reduction was not applied, so as to protect the speech signal. Interestingly, Synchrony Detection applied in isolation (i.e., without modulation detection) provides no information or decisions based on the presence or absenceof noise.

Maximizing Noise Reduction via Synergy:

Modulation and Synchrony detection analyze and respond to different aspects of the sound signal and they operate under different conditions.

The advantage of Synchrony Detection is the range of acoustic environments within which it provides benefit. The ability to identify the presence of speech via Synchrony Detection is uniquely effective at a S/N as low as 0 dB (Elberling, 2002). In contrast, when speech co-exists within a non-modulated background noise and the S/N becomes less than +10 to +15 dB (more challenging), the modulation detection system will have difficulty distinguishing between noise and speech-in-noise signals.

These two detection systems used in tandem maximize the unique advantages of each system (Flynn & Lunner, 2004). When combined, Synchrony Detection provides a binary decision regarding the presence or absence of speech, despite extremely adverse and challenging listening situations, while Modulation Detection provides information about the quantity of noise in each channel. With both detection systems active, given noise in the absence of speech, aggressive noise reduction occurs. When speech is present in noise, gain reduction is more limited, providing some relief from the noise without compromising critical speech information

Figure 4 provides an example of gain reduction applied in a combined system (Oticon Syncro). The left panel shows gain reduction applied when speech and noise were present. In this case, gain reduction is applied aggressively in the low and extreme high frequencies. In contrast, in the right panel, when speech was not detected, gain reduction was applied across the entire bandwidth of the device.

Noise Reduction Performance Analysis:

Since the goal of noise reduction systems is to improve the acceptability of hearing aids in noisy environments, an evaluation of the effectiveness should be focused on such a criterion. Ricketts & Hornsby (2005) evaluated a group of patients regarding preference judgments for modulation-based noise reduction with respect to noise-reduction-on versus noise-reduction-off across two levels of background noise and they evaluated directionality-active versus directionality-disabled. Across all conditions, patients showed a strong preference for noise reduction on.

Similarly, Powers et al. (2006) recently reported similar results when evaluating another modulation-based system. In blinded judgments after field use in a range of communication situations, patients showed an approximate 3 to 1 preference for noise reduction being active. The principle reason given by these patients was improved acceptability in noisy situations.

When noise reduction systems were evaluated with regard to the direct effect on speech understanding in noise, no effects were observed (Boymans et al., 1999). Because noise reduction systems do not change the S/N, speech improvements in noise are not expected, and are indeed, rare. Hearing aids without noise reduction can theoretically provide so much low frequency energy as to initiate an upward spread of masking (USOM). In such a case, gain reduction of the low frequencies secondary to noise reduction may minimize the USOM effect. However, most modern hearing aid fitting algorithms make specific attempts to minimize the USOM by carefully controlling audibility levels in the low frequencies.

Candidacy for Noise Reduction:

Nearly all patients experience highly variable acoustic environments with occasional loud and noisy sounds present throughout some portion of their day. It seems reasonable therefore, all patients should be considered candidates for noise reduction technology, unless a specific contraindication is present.

An ad hoc committee of the American Academy of Audiology (AAA) addressed noise reduction and amplification for children. The committee cautioned against noise reduction systems for younger children (Bentler et al., 2004) as they were concerned such systems might minimize speech audibility if noise reduction systems were applied too aggressively.

However, since current versions of these systems operate in a level dependent manner, it is unlikely important softer speech sounds would be rendered inaudible. Nonetheless, the concerns of the committee should be considered closely by clinicians fitting the youngest patients.

Sound Quality Preferences:

Indeed, some patients may not like the sound of the hearing aid changing in an otherwise stable sound environment. They may perceive noise reduction-based fluctuations as unnatural or distracting. Given that situation, the audiologist can disable the noise reduction system, assuming the manufacturer's hearing aid fitting software such an option. Disabling an otherwise valuable signal processing system is a decision which should only be made after the patient has sufficient time to acclimatize to the system. Clinical experience suggests that sometimes a particular advanced technology option may initially appear to be distracting, but may soon become transparent, unnoticed and ultimately acceptable and desired. Sometimes, noise reduction systems are adjustable. If a patient responds negatively to the noise reduction system as initially set, it can generally be adjusted to be more or less aggressive.

Figure 5 provides the output in KEMAR's ear canal of a MCNL hearing aid (Oticon Syncro) with environmentally adaptive directionality and noise reduction. Note, Syncro's directional system is designed to react to the environment more quickly than the noise reduction system. For these recordings, speech shaped noise was presented from 135 and 215 degrees. The Syncro was set in two different Identities (Identities are meta controls which impact a broad range of device systems in a coordinated manner, see Schum & Beck, 2006). The Calm identity (upper panel) is designed to respond slowly to changes in the environment while the Energetic identity (lower panel) responds to changes in the environment more aggressively.

The white arrow indicates the compression system responding to the application of speech noise at an intense level (75 dB SPL). The red arrow indicates the directional systems activation and reduced the noise coming from behind. Finally, the yellow arrow indicates the additional reduction in the speech noise as mediated by the application of noise reduction. Notice how all these events occur more quickly in the Energetic Identity and more slowly in the Calm Identity. Some patients may prefer more aggressive action whereas others may prefer a more restrained reaction or some setting in between these extremes. Currently, patient preferences are probably best determined via listening experience as opposed to a priori prediction.

Final Thoughts:

Successful use of noise reduction (and other advanced technologies) is most often dependent on the patient and family having realistic expectations. As is true and appropriate across many aspects of the professional relationship, the audiologist should counsel appropriately regarding appropriate and realistic outcomes from advanced technology hearing aids.

As may be the case with noise reduction, often it is more than just high expectations - it may be a case of a fundamental misunderstanding of what the system is designed to do. To facilitate a better understanding of available advanced hearing aid technology, manufacturers have developed multimedia counseling tools to help describe the benefit of advanced signal processing systems, such as noise reduction (Beck and McGuire, 2006).

References:

Beck, D.L., and McGuire, R. : MultiMedia: Better Tools Facilitate a Better Process. Hearing Review, May, 2006.

Bentler, R., Eiten, L., Gabbard, S., Grimes, A., Johnson, C., Moodie, S., Palmer, C., Ricketts, T. & Tharpe, A. (2004). Pediatric Amplification Guideline. Audiology Today. 16(2). Pp. 46-53.

Boymans M, Dreschler WA, Schoneveld PS, Verschuure H. (1999). Clinical evaluation of a full-digital in-the-ear hearing instrument. Audiology 38:99-108.

Dillon, H. (2001). Hearing Aids. New York: Thieme.

Elberling C. (2002). About the VoiceFinder. News From Oticon; January.

Flynn, M. C. (2003). Maximizing speech understanding and listening comfort in noise. The Hearing Review, 7, 50-53.

Flynn, M.C., & Lunner, T. (2004) Clinical evidence of the benefits of Oticon Syncro. News from Oticon, November.

Kryter, K. (1962). Mehtods for the calculation and use of the articulation Index. Journal of the Acoustical Society of America, 34:1689-1697.

Powers, T., Branda, E., Hernandez, A. & Pool, A. (2006). Study finds real-world benefit from digital noise reduction. The Hearing Journal, 59(20): 26-30.

Ricketts TA, Hornsby BWY. (2005). Sound quality measures for speech in noise through a commercial hearing aid implementing "digital noise reduction." Journal American Academy of Audiology 16:270-277.

Schum, D. (1996). Speech understanding in background noise. In M. Valente (Ed.), Hearing Aids: Standards, Options, and Limitations. New York: Thieme Medical Publishers, Inc.

Schum, D. (2001). Annoyance and hearing aids. Audiology Online, January 17, 2001. http://www.audiologyonline.com/audiology/newroot/articles/arc_disp.asp?id=247&catid=1

Schum, D. (2003). Noise-reduction circuitry in hearing aids: (2) Goals and current strategies. The Hearing Journal., 56(6), 32-41.

Schum, D, J. and Beck, D.L.: Meta Controls and Advanced Technology Amplification. 3-27-2006, see www.audiologyonline.com, http://www.audiologyonline.com/articles/article_detail.asp?article_id=1566