Sensorineural hearing loss has two components: sensitivity loss in quiet (attenuation) and processing loss in noise (distortion).1 There is a relationship between these two components, whereby as hearing loss in quiet increases, the ability to understand speech in noise decreases.2 This loss of speech understanding in noise is typically measured as the signal-to-noise ratio (SNR) required to maintain 50% correct understanding of speech in noise.
Performance measures collected with the Hearing-In-Noise Test (HINT)3 show normal-hearing listeners maintain 50% understanding at about -4 dB SNR, while for hearing-impaired listeners the SNR ability degrades by about 1 to 1.5 dB for every 10 dB of hearing loss in quiet.4,5 Thus, a person with a moderate sensorineural hearing loss may perform with about a 6 dB SNR loss compared to normal-hearing individuals.
The implication of this speech understanding deficit in noise is that hearing-impaired listeners need to be provided with a more favorable SNR in order to maintain near-normal speech intelligibility in background noise. In attempts to improve the SNR for hearing-impaired listeners, the use of directional microphones has been the only consistent method for providing hearing aid benefit in noise.6-8 Many studies with omni-directional hearing instruments, with or without digital noise reduction algorithms, have not shown significant benefit for tasks involving speech understanding in noise.7,9,10
These results have led to the conventional wisdom that directional microphones are the only proven method for improving the SNR for hearing-impaired listeners, and conversely, that multi-channel compression and digital noise reduction algorithms cannot improve the SNR. However, this assumption has been contended by recent studies demonstrating significant SNR benefit from a new omni-directional DSP hearing aid.11,12 The following laboratory study using the new Sonic Innovations Natura 2 SE directional instrument provides additional evidence for this contention.
Improving the SNR
Successful hearing instrument fittings must first restore the audibility of speech cues to compensate for hearing loss, without exceeding the individuals loudness tolerance levels. One accepted technique for accomplishing this is through digital multi-channel compression. Beyond compression, various signal processing approaches may be used to reduce background noise, thereby improving the SNR and compensating for SNR loss. Two common techniques designed to improve the SNR are directional microphones and digital noise reduction.
|Fig. 1. HINT sentence The boy fell from the window in spectrally matched, steady-state noise, as recorded through the Natura 2 SE BTE-D hearing instrument in four test conditions: a) Omni-directional (Omni) mode, which represents the digital multi-compression condition and serves as the reference recording (shown as the dark blue background in the rest of the figures); b) Omni plus noise-reduction (NR) mode showing the effect of incorporating digital noise reduction; c) Directional (Dir) mode showing the effect of changing the microphone condition, and d) Dir plus NR mode showing the combined effects of the directional microphone and noise reduction algorithm.|
Fig. 1 displays multiple time-waveforms of signals used in the HINT, recorded through the new digital directional BTE with noise reduction capability. The speech and noise were each presented at 65 dBA (creating a 0 dB SNR) in a quasi-free sound field (described below). The hearing instrument was set to 20 dB of gain and was attached to a Zwislocki four-branch-type occluded ear simulator. The instrument can be configured in four conditions as shown in Fig. 1.
In Fig. 1b (Omni+NR), the time course of the digital noise reduction algorithm can be seen as it progressively reduces the amplitude of the noise in the light foreground recording, but does not reduce the amplitude of the speech. In the directional condition (Fig. 1c), there is a small, but consistent, reduction in the noise amplitude without a significant reduction in the speech signal. The combined effects of noise reduction and directionality are seen on the speech and noise signals of Fig. 1d.
While the images in Fig. 1 show an improvement in the SNR at the device level, this does not necessarily mean that hearing-impaired listeners can benefit from the signal processing in the hearing instrument and reduce their SNR loss. The purpose of this study was to objectively measure speech understanding in noise in the four aided conditions and determine if there was any significant benefit with respect to the unaided condition.
The ISO Quasi-Free Field
A new test environment, with modified test materials, was created to evaluate the signal processing technologies in this study.13 Consideration was given to controlling both spatial and temporal variables in a way that recognized real-world factors, where possible.
Directional microphone hearing instruments require spatial separation between the speech and noise in order to show directional benefit. Prior laboratory research has shown that the laboratory benefit in directional microphone studies is contingent on the loudspeaker placement.14 Predictions of real-world benefit in laboratory measures are best obtained when a multiple loudspeaker array is used with equal spacing of the noise sources throughout the sound field.
|Fig. 2. Equipment design showing multi-track hard-disk recorder and the two-channel audiometer routed through a five-channel amplifier supporting a sound-field array consisting of five loudspeakers.|
A sound field array was created using five loudspeakers, each placed one meter from the subject (Fig. 2). With this multi-loudspeaker array, the masker coming from the RF, RR, LR, and LF speakers creates a uniform noise field with no front-to-back or right-to-left advantage. The sound field meets the ISO requirements for a quasi-free field.15
Using this multi-loudspeaker array, two different listening conditions were evaluated. In the noise front condition, the speech signal and masking noise are both delivered from the single loudspeaker in front of the subject. In the noise diffuse condition, the speech comes from the front speaker and the masker comes from the four surrounding loudspeakers.
For this study, the HINT materials were modified in three ways: 1) The presentation lists were increased from 10 sentences to 20 sentences to improve test reliability; 2) The onset of the masker before the sentences was lengthened from 0.5 seconds to 5.0 seconds to allow time for the digital noise reduction algorithm to engage; 3) For the noise-diffuse condition, the masker recordings presented from each loudspeaker were time-staggered with respect to each other to create an uncorrelated quasi-free noise field. These modifications to extend the noise leader and separate the masker sources were designed to represent more real-world listening conditions than the unmodified HINT.
An investigation was conducted at the manufacturers Hearing Aid Research Laboratory in Salt Lake City using the new NATURA® 2 SE directional BTE device. This instrument employs nine-channel compression with very fast and symmetric attack and release time constants. In normal clinical use, the omni-directional mode is paired with Speech Weighted Expansion (passive noise reduction) and the directional mode is paired with Personalized Noise Reduction (active noise reduction), creating what is called the DIRECTIONALplus system. The directional mode produces a hypercardioid pattern with an AI-DI of 4.6. The EXPRESSfit fitting system automatically configures the directional mode with a frequency response that is nearly equalized to that of the omni-directional mode.
Twenty adults, ages 34-84 years (mean=68 years), were enrolled in the study. Sixteen of the subjects were male and four were female. All subjects had bilateral, sensorineural mild-to-severe hearing loss (Fig. 3), and all but three had prior experience with amplification.
|Fig. 3. Mean and range of audiometric thresholds with ±1 standard deviation (n=20 subjects, 40 ears).|
The subjects were tested unaided with the HINT sentences in quiet (no masker) to verify that they had reception thresholds for sentences (RTS) of 55 dBA. The motivation for this criterion was to use only subjects who received a significant masking effect with 65 dBA of noise in the unaided condition (the standard presentation level for the HINT masker). This prevented benefit calculations from being artificially inflated when comparisons were made between the aided and unaided conditions.
Following qualification, the subjects were fit binaurally with the test hearing instrument following the manufacturers protocol based on the pure-tone audiogram combined with in-situ loudness mapping through the hearing instrument. Conventional, vented earmolds were used where possible. After the initial fitting, subjects wore the aids in their normal daily environments and returned for follow-up. Minor fine-tuning was performed as needed in response to subject comments on sound quality. Subjects were tested after the fittings were deemed satisfactory to the subject and the researcher. The average time between fitting and data collection was one month.
Speech intelligibility in noise measures were obtained in the quasi-free field with the modified HINT as described above. The testing sequence (Unaided, Omni, Omni+NR, Dir, and Dir+NR) and the listening environment (noise-front and noise-diffuse) were counterbalanced across subjects. Data were collected over two or three sessions to avoid repetition of HINT word lists within a session.
In a multivariate ANOVA comparing all aided conditions, there were significant main effects for the listening environment (noise diffuse = -1.19 dB SNR, noise front = +1.35 dB SNR) [F(1,19) 133.8, p<0.001], the noise reduction condition (NR on = -0.50 dB SNR, NR off = +0.66 dB SNR) [F(1,19)=7.77, p<0.001], and the microphone condition (directional = -0.68 dB SNR, omni-directional = +0.84 dB SNR [F(1,19)=54.1, p<0.001]. There was a significant interaction between the microphone condition and the test environment [F(1,19) = 33.5, p < .001]. In order to understand the interaction and relate performance to the unaided condition, the analysis was repeated for the noise-front and the noise-diffuse conditions, respectively (Fig. 4).
|Fig. 4. Mean HINT thresholds (in dB SNR) for two test environments (noise-front and noise-diffuse) in the unaided condition and four aided conditions (n=20 subjects).|
Noise-Front Condition: An ANOVA of the noise-front environment revealed a significant main effect of listening condition [F(4,76) = 16.3, p < .001]. Post-hoc analysis revealed significant differences between the unaided condition (4.4 dB SNR) [F(1,76)=13.8, p <.01] and all aided conditions. Comparison of omni-directional and directional results found no significant differences, while comparisons of results with and without noise reduction found significant differences [F(1,19) = 11.3, p < .01]. The differences below are described as benefit, obtained by comparing the unaided RTS to the aided RTS.
For the noise-front environment, there was a mean aided benefit of 2.6 dB SNR without NR activated and 3.5 dB SNR with NR activated. The significant effect of the NR condition is due to the algorithm exploiting the temporal modulation differences between the fluctuating speech and the steady-state noise. There was no significant difference between thresholds obtained in the omni-directional and directional modes in the noise-front condition. This was expected, as the omni-directional and directional modes have nearly identical frequency responses. The two modes should have equal sensitivity when there is no spatial separation between the speech and the noise.
Noise-Diffuse Condition: An ANOVA of the noise-diffuse environment also revealed a significant main effect for listening conditions [F(4,76) = 54.1, p < .001]. Post-hoc analysis revealed significant differences between the unaided condition (3.1 dB SNR) and all aided conditions [F(1,19) = 21.8, p < .01]. For the individual aided conditions, the differences below are described as benefit, obtained by comparing the unaided RTS to the aided RTS.
|Omni vs. Omni + Noise Reduction|
|Omni + NR||3.5|
|Omni + Noise Reduction vs. Directional|
|Omni + NR||3.5||17.3||<0.01|
|Directional vs. Directional + Noise Reduction|
|Dir + NR||6.5|
|Table 1. ANOVA results comparing noise-diffuse conditions. The dB SNR values represent benefit compared to unaided conditions.|
Table 1 shows that the benefit obtained in the Omni+NR condition (3.5 dB SNR) was greater than the Omni condition (2.5 dB SNR). Benefit in the Dir condition (4.8 dB SNR) was greater than the Omni+NR condition, and benefit in the Dir+NR condition (6.5 dB SNR) was greater than the Dir condition.
The best performance in the quasi-free field was obtained in the Dir+NR condition. This benefit is thought to result from a combination of the core signal processing with digital multi-channel compression (providing audibility of speech cues), the noise reduction algorithm (using modulation differences to improve the SNR), and the directional microphones (using spatial separation to improve the SNR). The total aided benefit of 6.5 dB SNR significantly exceeds the AI-DI for these hearing instruments, which is an exceptional finding for measurements obtained in a quasi-free noise field.
|Fig. 5. Scatter-plot with linear regression lines of unaided vs. aided HINT RTS (dB SNR) for four aided conditions in the quasi-free sound field (n=20 subjects). The dotted, diagonal line indicates equal performance in noise for the unaided and aided conditions. Data points below this line indicate better performance in the aided condition.|
Fig. 5 is a scatter-plot of the unaided thresholds in noise (referenced to the abscissa) vs. the aided thresholds in noise (referenced to the ordinate) for the four amplification conditions evaluated in the quasi-free field. The dotted diagonal line represents equal speech understanding in noise for the unaided and aided conditions. Data points above the dotted line indicate better performance in the unaided condition and points below the dotted line represent better performance in the aided conditions. For this experiment, 85% (17 of 20 subjects) of the Omni data points are below the dotted line, 95% (19/20) of the Omni+NR and Dir data points are below the dotted line, and 100% (20/20) of the Dir+NR data points are below the dotted line.
Linear regression lines are plotted in Fig. 5 for each of the four aided conditions. The distance between each regression line and the dotted diagonal line indicates the expected aided benefit for the specific listening condition. Progressive benefit is seen as signal processing technologies are incorporated for the Omni, Omni+NR, Dir, and Dir+NR conditions. For example, an average subject with an unaided threshold of +2 dB SNR could be reasonably expected to hear at about 0 dB SNR in the omni-directional condition, at about -1 dB SNR for omni-directional with noise reduction, at about -2 dB SNR in the directional condition, and about -3 dB SNR for directionality with noise reduction.
If the regression lines were parallel to the dotted line, subjects would be obtaining the same amount of benefit without regard to their unaided threshold in noise. The fact that the regression lines are canted horizontally indicates that the subjects with greater SNR loss are receiving more benefit for speech understanding in noise than those subjects with less SNR loss. These results reveal that persons with the most trouble understanding speech in noise should obtain the most benefit from the test hearing aids.
Statistical analysis of the noise-front and noise-diffuse environments demonstrate significant benefits for speech intelligibility in noise in all aided conditions: omni-directional or directional, with or without noise reduction. These results challenge the conventional wisdom that only directional microphones can improve the SNR performance for hearing aid wearers, and that multi-channel compression and digital noise reduction algorithms do not yield improvements in objective SNR tests.
The data demonstrate a reduction in SNR loss for hearing-impaired subjects from the individual factors of a) the core signal processing (mean of 2.4 dB SNR seen across the noise-front and the noise-diffuse conditions); b) noise reduction (mean of 1.2 dB SNR across the noise-front and noise-diffuse conditions), and c) directionality (mean of 2.6 dB SNR from the noise-diffuse condition).
In addition, the data show for the first time an additive effect of the three signal processing technologies (2-3 dB SNR for the core signal processing, 1 dB SNR for the noise reduction, and 2-3dB SNR for directionality), producing a combined effect of 6 dB SNR with the DIRECTIONALplus configuration, which represents a greater benefit than any individual signal processing feature for the instrument.
This article was submitted to HR by Victor Bray, PhD, & Michael Nilsson, PhD, of the Auditory Research Dept. of Sonic Innovations, Inc., Salt Lake City. Correspondence can be addressed to HR or Victor Bray, PhD, Sonic Innovations, Inc., 2795 East Cottonwood Parkway, Suite 100, Salt Lake City, UT 84121-7036; email: [email protected].
The authors thank Robert M. Ghent, Jr., Patrick J. Murphy, Rebecca Nilsson, and Roxanne Olsen for their help in this project.
1. Plomp R: Auditory handicap of hearing impairment and the limited benefit of hearing aids. J Acoust Soc Amer 1978; 63: 533-549.
2. Killion M: Hearing aids: Past, present, future: Moving toward normal conversations in noise. Brit Jour Audiol 1997; 31: 141-148.
3. 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. JASA 1994; 95(2): 1085-1099.
4. Nilsson M, Bray V, Ghent R, & Murphy P: Single microphone noise reduction: New findings. Research poster at the Amer Acad of Audiol Convention, San Diego, 2001.
5. Bray V & Nilsson M: Assessing hearing aid fittings: An outcome measures battery approach. In M Valentes Strategies for Selecting and Verifying Hearing Aid Fittings, 2nd Edition. New York: Thieme, In press.
6. Hawkins D & Yacullo W: Signal-to-noise ratio advantage of binaural hearing aids and directional microphones under different levels of reverberation. J Speech and Hear Disorders 1984; 49: 278-286.
7. Ricketts T & Dhar S: Comparison of performance across three directional hearing aids. J Amer Acad Audiol 1999; 10: 180-189.
8. Valente M, Sweetow R, Potts L & Bingea B: Digital versus analog signal processing: Effect of directional microphone. J Amer Acad Audiol 1999; 10: 133-150.
9. Knebel S & Bentler R: Comparison of two digital hearing aids. Ear & Hear 1998; 19(4): 280-289.
10. Walden B, Surr R, Cord M, Edwards B & Olson L: Comparison of benefits provided by different hearing aid technologies. J Amer Acad Audiol 2000; 11: 540-560.
11. Bray, V & Nilsson, M: Objective test results support benefits of a DSP noise reduction system. Hearing Review 2000; 7(11): 60-65.
12. Bray V & Valente M: Can omni-directional hearing aids improve speech understanding in noise? Audiology Online, Sept 24, 2001.
13. Nilsson M, Ghent R, Bray V & Enrietto J: Development of test materials to evaluate noise reduction and directional hearing aids. Manuscript submitted for publication, 2001.
14. Ricketts T: Impact of noise source configuration on directional hearing aid benefit and performance. Ear & Hear 2000; 21(3): 194-205.
15. International Organization for Standardization: AcousticsAudiometric test methods, Part 2: Sound field audiometry with pure tone and narrow-band test signals. 1992, ISO 8253-2.