Adults with mild and moderate hearing loss are not necessarily motivated to purchase hearing instruments because they cannot hear, but because they cannot hear in noise. Despite this need, hearing instruments have historically offered only limited benefit in background noise.1,2,3

Many dispensing professionals address the noise problem by fitting directional hearing instruments that have been shown to help improve speech intelligibility in noise.4,5 These improvements can be demonstrated provided that certain criteria are met, in particular that the speech and noise are spatially separated. While extremely useful in many situations, directional microphones provide limited or no benefit when the speech and noise originate from the same location, as often occurs in real life. Benefit from directional hearing instruments also decreases as listening conditions change from anechoic to more reverberant environments.6,7 

Another clinical approach for overcoming background noise is to fit hearing instruments with noise reduction circuits. Modern algorithms implemented in digital devices generally use a method called “spectral subtraction” whereby the spectrum of competing noise is estimated and subtracted from the total signal. In this process, noise is identified as any steady-state waveform, whereas speech is characterized as any modulated, or time-varying waveform. These algorithms act to selectively reduce gain in frequency regions containing noise, but not those regions containing speech. The effect is to preferentially amplify the portions of the total signal with better speech-to-noise ratios. 

Electroacoustic measurements of noise reduction hearing instruments predict that improved speech-to-noise ratios are possible. To date, evaluations on subjects have shown improved listening comfort in background noise, but no systematic improvement in speech intelligibility in noise.8,9,10 Traditional implementations of spectral subtraction may sacrifice benefit to prevent the acoustic artifacts commonly generated by this technique. Some of these artifacts can be objectionable and interfere with use of the device in certain listening situations.

Using patented digital signal processing (DSP) techniques, Sonic Innovations has developed hearing instruments designed to help most people hear better in a variety of noisy environments. The new Natura 2 SE and Conforma 2 SE hearing instruments include a noise reduction algorithm based on spectral subtraction, as described above. Combining this noise reduction with a single, omni-directional microphone and fast, narrow-band compression, the instrument is designed to avoid many of the limitations often associated with directional microphones and other noise reduction algorithms.

The Noise Reduction System

In designing the DSP devices for Sonic Innovations, several basic assumptions or engineering “rules” about hearing instrument amplification were broken for the purpose of creating an improved hearing system.11,12 Signals are processed in nine independent channels, each working with very fast and symmetric temporal characteristics. Additionally, the time constants are frequency dependent, with faster attack and release times for higher frequencies. Compression is applied to the signal envelope, leaving the spectral content unchanged. The compression channels are narrow, operating at half-octave bandwidths, and are aligned with conventional audiometric test frequencies.

The noise reduction system begins with a noise detector that analyzes the real-time auditory spectrum by seeking steady-state signals in each channel. Once the noise estimate is determined, the DSP establishes a “target”-to-“masker” relationship, or speech-to-noise ratio (S/N), in each compression channel. The noise reduction algorithm attenuates portions of the input, depending upon the S/N, with greater attenuation as the S/N decreases.

Fig. 1. Speech in noise with (green) and without (blue) noise reduction.

Fig. 2. Mean audiogram for 27 subjects, separated by device style.

Fig. 3. Mean real ear insertion responses (REIR) for 46 custom aids.

The system provides multi-channel amplification of modulated signals and a reduction in steady-state maskers. Fig. 1 shows an unprocessed waveform for a nine-second recording of a male talker saying “bathe, it, wet, give,” mixed with continuous white noise at +5 S/N. Overlaid is a processed waveform of the same mixed signals demonstrating several characteristics of the system. First, the noise reduction engages over time, beginning two seconds after the noise onset, and reaching maximum effectiveness five seconds post-onset. Second, the attenuation created by the noise reduction system occurs when and where (i.e., in the time and frequency domain, respectively) speech is not present. As seen in Fig. 1, the amplitude envelope of the speech (a modulated signal) is not reduced.

This noise reduction technique is called Personalized Noise Reduction™ for several reasons. First, there are three programmable levels of noise reduction that are audiogram-based. The noise reduction levels of low, medium and high correspond to 6, 12 and 18 dB maximum noise reduction, respectively. Noise reduction is applied before the AGC, with the level of the reduction dependent on the amount of gain programmed in any given compression channel. This “personalization”—or dependence on the individual’s hearing loss and gain requirements—attenuates the noise to a level near the individual’s threshold regardless of the severity of the hearing loss.

Clinical Study

A clinical research study was undertaken to evaluate if the above combination of core signal processing with noise reduction improves speech comprehension in noise. Benefit is expressed as a change in performance between unaided and aided conditions, and between aided conditions with and without noise reduction. 

Fig. 4. Mean HINT scores for 27 subjects. Lower thresholds indicate better scores.

Fig. 5. Mean APHAB scores for 27 subjects.

Test conditions reduced or eliminated spectral, temporal and spatial cues. Noise with the same long-term-average spectrum as the sentences eliminates cues from spectral mismatch. Unmodulated or steady-state noise removes any temporal gaps to make the task relate to “speech detection in noise” rather than “speech detection in noise gaps.” A single loudspeaker for both target and masker removes the spatial cues that occur when speech and noise originate from different azimuths.

Methods: Twenty-seven hearing-impaired adults were fit binaurally with CIC, ITC and ITE devices. The audiogram in Fig. 2 shows mean, minimum and maximum thresholds across frequencies. A loudness normalization fitting technique with in-situ verification of dynamic range generated initial fittings. Fitting adjustments were made, as needed, to address patient preferences related to occlusion, loudness, timbre, etc. 

Real-ear insertion response, with and without noise reduction, quantified the hearing instrument gain, as well as the change in insertion response with noise reduction engaged. Short (<2 sec.) and long (²10 sec.) presentations of speech-weighted composite noise were used.

Fig. 3 shows the average insertion responses across 46 ears for each level of noise reduction (none, low, medium, high). The maximum level of noise reduction varied across frequencies appropriate to the differences in prescribed gain across frequencies. The graph demonstrates that there is no change in the frequency response for short stimuli with any level of noise reduction, but there is a reduction in the frequency response up to 6, 12 and 18 dB when signal duration exceeds the onset time of the noise detector. 

Speech intelligibility was measured in the soundfield using a single loudspeaker placed one meter in front of the subject at ear level. Subjects were tested unaided and aided, with and without noise reduction, using the Hearing in Noise Test (HINT).13 HINT noise was presented at the higher value of 65 dB(A) or 20 dB SL (relative to thresholds in quiet). Because of the 5-sec. onset of the noise reduction algorithm, the HINT masker was modified to extend the noise onset-time from 0.5 sec. to 5 sec.

HINT scores are expressed as a Reception Threshold for Sentences (RTS) and correspond to the presentation level (relative to the noise level) at which sentences can be repeated correctly 50% of the time. The scores are expressed in dB S/N, with a lower threshold associated with a better score (i.e., performance can be maintained in a more difficult listening condition).

Results: Analysis of variance of RTS shows a significant benefit of amplification and noise reduction [F(3,72)=20.32, p<.001]. First, amplification significantly lowered RTS versus thresholds in the unaided conditions. Second, any level of noise reduction significantly lowered thresholds relative to the aided conditions without noise reduction (Fig. 4). It is possible to interpret this change in RTS as a corresponding change in intelligibility. The performance intensity function of the HINT materials has a slope of almost 10% for each dB change in RTS.13 This means that the 3.3-3.6 dB difference between the unaided and aided-with-noise-reduction conditions corresponds to approximately 35% improvement in intelligibility in noise using amplification with noise reduction.

Subjective benefit was evaluated with the APHAB14 for reference devices (original Natura) and devices with noise reduction (Natura 2 SE). A significant main effect was found [F(1,21)=5.34, p<.05] with more benefit on all APHAB subscales for the devices with noise reduction (see Fig. 5).


Significant improvements in speech intelligibility in noise were obtained using DSP amplification combined with a new noise reduction algorithm. This benefit was achieved in a difficult listening situation where there was no spatial separation or spectral mismatch between the speech and noise, nor temporal gaps in the noise. 

In contrast to other studies of noise reduction, objective measures support the subjective ratings. Subjects who already demonstrated high benefit scores with reference devices (i.e., original Natura) showed increased benefit with the noise reduction devices. This increase in benefit was shown in all APHAB subscales, including reverberation and background noise conditions. 

The data show that, contrary to conventional belief, most listeners with mild and moderate sensorineural hearing loss can achieve significant improvements in speech understanding in noise using an omni-directional microphone configuration combined with the noise reduction signal processing described in this article.

The findings imply that clinicians can effectively combine the known benefits of CIC and ITC instruments (e.g., improved cosmetics, reduced occlusion, reduced wind noise, improved localization) with new signal processing and obtain high levels of user satisfaction. In addition, the system reported here is designed to remain effective in conditions where directional hearing instruments, by definition, cannot show benefit—namely where speech and noise originate from the same source. w


The field studies were conducted under the supervision of Michael Nilsson, PhD, with contributions from Bob Ghent, MS, Patrick Murphy, MA, Merritt Johns, MS, Rebecca Nilsson, BA, and Roxanne Olsen, CDWH.


1. Kochkin S: MarkeTrak III identifies key factors in determining customer satisfaction. Hear Jour 1993; 46 (8): 39-44.

2. Kochkin S: Customer satisfaction and subjective benefit with high-performance hearing aids. Hearing Review 1996; 3 (12): 16-26. 

3. Kochkin S: MarkeTrak V: “Why my hearing aids are in the drawer”: The consumer’s perspective. Hear Jour 2000; 53 (2): 34-42. 

4. Valente M, Fabry D & Potts L: Recognition of speech in noise with hearing aids using dual microphones. J Amer Acad Audiol (JAAA) 1995; 6 (6): 440-449.

5. Wouters J, Litiere L & van Wieringen A: Speech intelligibility in noisy environments with one- and two-microphone hearing aids. Audiology 1999; 38 (2): 91-98.

6. Ricketts T & Dhar S: Comparison of performance across three directional hearing aids. JAAA 1999; 10 (4): 180-189.

7. Ricketts T: Impact of noise source configuration on directional hearing aid benefit and performance. Ear Hear 2000; 21 (3): 194-205.

8. Valente M, Fabry D, Potts L & Sandlin R: Comparing the performance of the Widex Senso digital hearing aid with analog hearing aids. JAAA 1998; 9 (5): 342-360.

9. Boymans M, Dreschler W, Shoneveld P & Verschuure H: Clinical evaluation in a full-digital in-the-ear hearing instrument. Audiology 1999; 38 (2): 99-108. 

10. Versfeld N, Scheyen D, Sidonne C, van Kreveld-Bos G, Hendricks J & Dreschler W: Field trial with the Danavox Danalogic hearing aid. Presentation to the 25th International Congress of Audiology, The Hague, The Netherlands, 2000. 

11. Bray V, Nilsson M, Ghent R & Johnson J: Results from a clinical evaluation of a new DSP hearing aid using an optimized platform. Presentation to the 11th Amer Acad of Audiol (AAA) Convention, Miami Beach, FL, 1999.

12. Bray V, Nilsson M, Ghent R & Murphy P: A new approach to speech intelligibility using DSP hearing aids. Presentation at the Brit Soc of Hearing Aid Audiol 2000, Brighton, United Kingdom, 2000.

13. Nilsson M, Soli S & Sullivan J: Development of the Hearing In Noise Test for the measurement of speech reception thresholds in quiet and in noise. J Acoust Soc Amer 1994; 95 (2): 1085-1099.

14. Cox R & Alexander G: The Abbreviated Profile of Hearing Aid Benefit. Ear Hear 1995; 16 (2): 176-186.