This article offers insight into the development of a new digital noise reduction algorithm, Voice iQ. Data are presented documenting patient preference for this fast-acting algorithm during real-world listening.

As with numerous algorithms in modern hearing aids, digital noise reduction (DNR) can be implemented through the use of fast or slow time constants. The function of DNR algorithms is to reduce prescribed gain in response to high levels of environmental noise.

William S. Woods, PhD Nazanin Nooraei,  AuD Jason A. Galster, PhD Brent Edwards, PhD
William S. Woods, PhD, is a principal research scientist and Nazanin Nooraei, AuD, is a research audiologist at the Starkey Hearing Research Center, Berkeley, Calif; Jason A. Galster, PhD, is manager of clinical comparative research at Starkey Laboratories, Eden Prairie, Minn; and Brent Edwards, PhD, is vice president of research at the Starkey Hearing Research Center. Correspondence can be addressed to HR or William Woods at .

The time constants of these systems control how quickly gain is changed in response to acoustic changes around the patient. Systems with slow time constants estimate a long-term signal-to-noise ratio (SNR) and reduce gain in a gradual manner, over seconds or hundreds of milliseconds. This method has been shown to improve sound quality and comfort.1,2

In contrast, DNR with fast time constants will manage the reduction and application of gain very quickly, over milliseconds. A challenge to the design of any DNR algorithm is the fact that speech quality may be compromised, for instance, if gain is reduced in the presence of noise and not accurately reapplied when speech is the dominant signal; this may result in distortion of the speech. With slow-acting DNR, the result is a system that must be inactive when speech is the dominant signal, with the disadvantage of not reducing the perception of noise.

Toward a More Intelligent System

Realizing the limitations of the slow-acting DNR—and the possibility of overcoming this limitation—the Starkey Hearing Research Center developed Voice iQ, a fast-acting DNR algorithm designed for the suppression of noise in the presence of an acoustically dominant speech signal. With this design, Voice iQ analyzes and classifies speech and noise: when speech is the dominant signal, prescribed gain is applied; when noise is the dominant signal, gain is reduced. The key is that the analysis occurs at a rate that allows audible noise between speech segments to be momentarily reduced, resulting in the perceived reduction of ambient noise and the preservation of the speech signal.

Prior to the implementation of Voice iQ, it was important to establish the efficacy of an advanced noise reduction algorithm by measuring the subjective impressions/perceived impact in real-world environments. This question was investigated by asking participants with hearing loss to wear hearing aids with both slow-acting and fast-acting noise reduction systems in real-world listening environments. It was also of interest to determine whether degree of hearing loss impacted the preference.

Research on Optimization of Digital Noise Reduction

FIGURE 1. Mean audiograms for participants in the mild and moderate hearing loss groups.

To determine the parameters of the fast noise reduction algorithm used in this study, laboratory tests compared hundreds of different parameter settings for the key elements integral in a noise reduction algorithm: time constants, amount of gain suppression, and frequency resolution. This optimized version of the algorithm was judged by study participants to be as good as or better than similar algorithms from the engineering literature, requiring the processing capabilities of a desktop computer.3

The findings of this study, reported at the International Hearing Aid Research Conference in 2006, resulted in a fast-acting DNR system optimized for the needs of the hearing-impaired patient.

Study subjects and methods. A total of 12 hearing-impaired individuals, 6 with mild and 6 with moderate hearing loss, participated in this study. Participants ranged in age from 60 to 84, with a mean age of 72. All participants had symmetric, sensorineural, sloping hearing loss, and 4 were experienced hearing aid users. Figure 1 shows the mean audiometric data of the participants.

For a total of 6 days, participants wore bilateral BTE hearing aids with slow-acting, multi-band WDRC initially fit to NAL-NL1 prescriptive targets. Adjustments to gain and compression parameters were made to provide a comfortable fit. The hearing aids were programmed with two memories: the first with the optimized fast DNR algorithm and the second with a slow-acting DNR. Only the DNR varied across memory, and the memories were counterbalanced across participants. Participants were blinded as to what the memories contained. No other processing (eg, directional processing, feedback cancellation) was active in the hearing aids.

Prior to participating in the study, participants were trained on changing the memories in the hearing aids and logging data to the hearing aid by holding the memory button down. This manual datalogging technique allowed the hearing aid to sample the acoustic environment at the command of the participant. Participants were instructed to go about their normal daily routines, comparing the two memories in various environments and recording their experience in a written diary. Participants were asked to note the preferred memory, date, time, location, sound level (eg, quiet, moderate, loud), and sound type (eg, car, restaurant, voices). They were asked to trigger the manual datalogging function when making a diary entry in order to accurately sample the acoustics of the real-world environment. These data were synchronized with the written diary data by using a time stamp in the logged hearing aid data.

FIGURE 2. Preference ratings for fast and slow digital noise reduction for groups with mild and moderate hearing loss.

Results. A total of 58 preference ratings were collected from the group with mild loss, and 46 ratings were collected from the group with moderate hearing loss. The diary entries indicated that many different environments were sampled, representing typical quiet and loud, indoor and outdoor real-world scenarios. Some of these included locations such as homes, churches, stores, theatres, and restaurants. Sources of noise were identified as voices, TV, music, traffic, and household appliances.

Preference data was collapsed across all of these environments and separated by hearing loss group. These results, shown in Figure 2, demonstrate that both groups preferred the fast algorithm parameters. A X2 test of homogeneity showed that, while both groups preferred fast-acting DNR, the mild-loss group showed a significantly stronger preference than the moderate-loss group [X2 = 14.84, X2(a=0.01, df=2) = 9.21]. Analysis of the logged variables (such as amount of gain reduction or long-term SNR) and subjective descriptions of environments in the diaries (such as overall loudness) did not yield distinct differences between the two groups; however, these qualitative results support the finding that both groups preferred the fast-acting noise reduction to the slow-acting. Some of the field comments included:

“The background noise was less busy [with fast-acting noise reduction].”
“Sitting in the backseat, voices seemed clearer against the hum of the motor [with fast-acting noise reduction].”
“Traffic and wind noise are much quieter [with fast-acting noise reduction].”


Based on the results of this single blinded, multi-day field trial, listeners with mild and moderate hearing loss significantly preferred the optimized fast-acting noise reduction strategy when compared to a slow-acting strategy in real-world listening situations. Specifically, the strength of preference was greater for participants with mild hearing loss. This outcome was later used to establish adjustable parameters for the final implementation of the noise reduction algorithm.

The successful outcome of this study supported the decision to move this algorithm from a research phase to a product development phase. This process of systematic validation throughout research and development allowed a newly developed digital noise reduction algorithm to become Voice iQ, Starkey’s currently available fast-acting DNR system offered in the S Series iQ hearing aids.


Data cited in this article were originally presented at the International Hearing Aid Research Conference held in Lake Tahoe during August 2006.


  1. Bentler R, Chiou LK. Digital noise reduction: an overview. Trends Amplif. 2006;10(2):67-82.
  2. 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.
  3. Martin R, Malah D, Cox RV, Accardi AJ. A noise reduction preprocessor for mobile voice communication. EURASIP J Applied Signal Processing. 2004;8:1046-1058.

Citation for this article:

Woods WS, Nooraei N, Glaster JA, Edwards B. Real-world listening preference for an optimized digital noise reduction algorithm. Hearing Review 2010;17(9):38-43.