Improving Speech Recognition in Noise Using Binaural Beamforming in ITC and CIC Hearing Aids
By Matthias Froehlich, PhD, and Thomas A. Powers, PhD
Directional microphone technology for hearing aids was introduced in the late 1960s in Germany, a few years later in the United States, and for decades has been considered the best solution using an ear-worn instrument to improve speech understanding in background noise. For some patients, however, the use of directional technology has presented a style selection dilemma. If small and discreet instruments (eg, ITCs or CICs) are a high priority, and this form factor is preferred by the patient, until recently, effective directional processing was not possible. The patient, or the hearing care professional, would then have to decide what is most important—least visible hearing aids or better speech understanding in background noise? In the past year, however, through the use of new beamforming technology, a solution to this long-standing quandary has become available.
While originally in BTE products, directional microphones have been used in custom instruments (full-concha ITEs) since the mid-1970s.1,2 In later years, when dual microphone processing became the norm for directional technology, researchers compared the directional benefit for ITEs versus BTEs.3,4 Their findings revealed that the benefit obtained with directional ITE instruments was similar to that of their BTE counterparts, and the absolute performance (speech recognition in background noise) was essentially the same.
Directional technology is most effective in custom instruments when the faceplate is flush with the exterior rim of the concha—in other words, a large full-concha ITE. This style typically is not very cosmetically appealing, and goes against the very reason why most patients want custom instruments. In the past, there have been several issues preventing the successful implementation of directional technology in the smaller ITC and CIC styles:
- There must be room within the instrument itself for the microphones and circuitry required for the directional processing.
- There must be effective port spacing to achieve the desired directivity. Consider that, in a CIC instrument, the spacing can be no greater than the diameter of the ear canal (about 7 mm), with no more than 4-5 mm to work with after the dimensions of the exterior shell are subtracted.
- The third issue to consider when designing directional processing in small custom instruments is that the deeper the faceplate lies within the concha, the more the attenuation and reflections of sound caused by the pinna disrupt the timing and amplitude differences of the sound entering the two ports—sounds from the front could strike the back port first. It is the precise analysis of the timing between ports that makes directional hearing aids directional.
Given all the factors working against effective directionality in small custom instruments, it’s probably not surprising that these products have not been available. This is changing, however, thanks to recent advancements in signal processing. The most recent development regarding directional hearing aid processing involves wireless communication between hearing aids, which allows for the exchange of audio data received by the microphones of both the right and left instruments, which then can be used to achieve narrow beamforming. This new beamforming technology from Siemens, termed Narrow Directionality, was described in detail by Kamkar-Parsi et al,5 and clinical research with this technology has revealed significant benefit for speech recognition in background noise, even when compared to individuals with normal hearing (for details, see Froehlich et al6 and Powers and Froehlich7).
Binaural Beamforming in ITCs and CICs
Following the success of the initial research, the next logical step was to take the clinically proven binaural beamforming approach to smaller custom instruments, where until recently, effective directionality was not available. With ITC products, it is still possible to use two inlet ports, and therefore, the beamforming algorithm operates much like it does with the BTE and larger ITE models. However, with the CIC instrument, where space allows for only one inlet port, the directionality task was more challenging.
The solution was binaural OneMic directionality (see Herbig and Froehlich8for review). With this approach, each hearing device in a bilateral pair operates with the exchange of audio signals detected and processed by both hearing aids. It is possible to then design a binaural adaptive beamforming algorithm that incorporates head-shadowing effects by carefully weighting and combining both available microphone signals, and by imposing an appropriate optimization criterion for the adaptive weighting rule. This results in an enhanced output signal, where interfering lateral noise sources can be efficiently attenuated while the frontal desired speaker signal remains untouched.8
Measures of sAI-DI
Previous papers have explained how the effectiveness of this beamforming technology can be evaluated, and product comparisons can be made through the measure of the sequential articulation-index-weighted directivity index, or sAI-DI.8-10 These sAI-DI measures are based on the previous work of Hagerman and Olufsson,11 and extends the approach suggested by Wu and Bentler.12 It is described in detail in a recent publication of Aubreville and Petrausch.13
Using the sAI-DI laboratory approach, it was possible to compare the newly designed ITC and CIC beamforming instruments to the previously researched mini-BTE/RIC models. These findings revealed an sAI-DI of 10.5 dB for the ITC, slightly higher than the 9.2 dB for the RIC instruments. As expected, because of factors discussed earlier, the sAI-DI for the CIC OneMic design was somewhat smaller at 5.1 dB, but still considerably superior to the sAI-DI of -0.5 dB for the traditional CIC omnidirectional.
While the laboratory sAI-DI measures were encouraging and provide an excellent method to make comparisons among products, there is only an indirect relationship between these values and SNR improvement for speech recognition in background noise. It was therefore also necessary to conduct clinical behavioral measures to assess the efficacy of these instruments. This testing was conducted at two different independent sites.
Clinical Evaluation of ITC Binaural Beamforming
Study Methods. Research examining the ITC instruments was conducted at Vanderbilt University in Nashville. The hearing-impaired participants were 20 individuals with bilateral mild-moderate downward-sloping sensory/neural hearing loss. Mean age was 66.6, with a range of 49-76 years. A group of normal-hearing individuals also were tested (unaided). Their mean age was 60.8 (range=53-71 years).
The hearing aids used in this research were the Siemens Insio 7bx custom-made ITCs. The gain and output of the hearing aids were programmed for each individual based on their hearing thresholds, according to the Siemens proprietary fitting algorithm (referred to as “binax fit”). Aided testing, with the participants fitted bilaterally, was conducted for three different programmed settings in the same instruments: omnidirectional, conventional directional, and frontal beamforming (Narrow Directionality). All other special features, such as digital noise reduction and feedback suppression, were activated and set to the manufacturer’s default settings.
The design for the speech recognition measures were the same as those used in earlier research with the Siemens mini-BTE/RIC instruments (see Froehlich et al 2015 for review).6,7 The target speech material was the sentences of the Hearing In Noise Test (HINT).14 The competing signal also was the HINT sentences, with the gaps between sentences removed, presented uncorrelated from the seven loudspeakers surrounding the listener. A speech babble was added to the competing HINT sentences, presented 15 dB below the sentence material.
Testing was conducted in a standard audiometric test suite. The array for the presentation of the target and competing speech material consisted of eight loudspeakers surrounding the participant, equally spaced at 45° increments, starting at 0° (ie, 45°, 90°, 135°, etc). The participant was seated 1.5 meters from all loud speakers in the center of the room, directly facing the speaker at 0°, which was used to present the target sentences. The competing signals were presented from the seven other loudspeakers; the resulting competing signal was 68 dBA at the position of the participant. The competing noise remained constant and the sentences were presented adaptively in the conventional manner, two HINT lists for each condition. The conventional HINT scoring method for 20 items was applied to calculate reference threshold for sentences (RTS, or SNR-50).
Results. The results of the study, illustrating the benefit provided by the binaural beamforming algorithm, are shown in Figure 1. Speech recognition using conventional directional processing also is shown for comparison. The benefit illustrated is the difference between the mean aided performance for each directional feature compared to the mean aided performance for the omnidirectional setting. As shown, benefit for traditional directional (2.3 dB) was substantially better than omnidirectional, and the binaural beamforming benefit (4.0 dB) was superior to traditional directional. This beamforming benefit with ITC instruments (compared to omnidirectional) is similar to that reported in previous research for the Siemens mini-BTE/RICs.
Statistical analysis of the speech recognition in noise data was completed using a generalized linear model for the three microphone settings (omnidirectional, adaptive directional, and beamformer). Results revealed a significant main effect of microphone setting (F = 48.44, p < 0.001, partial η2 = 0.566). No other significant main effects or interactions were present. Follow-up testing with Bonferroni correction indicated all settings were significantly different from each other, with the best performance from the beamformer (p < 0.003) and the next best performance from the directional setting (P< 0.001).
As mentioned, testing also was conducted for individuals with normal hearing. These comparisons are shown in Figure 2, expressed as the mean HINT SNR-50 for each condition. Analysis of these speech recognition in noise performance data was completed using a generalized linear mixed model comparing the three microphone settings (omnidirectional, adaptive directional, beamformer) to the performance of the group of normal-hearing listeners. This analysis indicated that the normal-hearing listeners’ performance was not significantly different than the hearing aid wearers using the adaptive directional setting, but was significantly poorer than those same hearing aid wearers when they were fitted with the beamformer technology (P< 0.038). As shown in Figure 2, when the binaural beamforming feature was enabled, the hearing-impaired group outperformed the normal-hearing group by 2.2 dB; again, this finding is similar to that reported by Froehlich et al6 for the mini-BTE/RIC beamforming instruments.
Clinical Evaluation of CIC Binaural Beamforming
The clinical trials for evaluation of the CIC binaural beamforming OneMic algorithm was conducted at the Hörzentrum Oldenburg, Germany. Participants were 15 individuals with mild-to-moderate bilateral downward-sloping sensorineural hearing loss. The mean age was 63.7 with a range of 51 to 74 years. The hearing aids used in this research were the Siemens Insio 7bx custom-made CICs; participants were fitted bilaterally for aided testing. The gain and output of the hearing aids were programmed for each individual based on their hearing thresholds, according to the Siemens “binax fit.” Testing was conducted for two programmed settings: omnidirectional and binaural OneMic beamforming. All other special features were activated and set to the manufacturer’s default settings.
The target speech material used was the sentences of the Oldenburger Satztest, commonly referred to as the OLSA.15 Similar to the design of the Vanderbilt ITC research, the background competing signal also was the OLSA sentences, presented uncorrelated from four different loudspeakers surrounding the listener, with the gaps between sentences removed, and a speech babble added at a level 15 dB below the sentences. The target speech signal was presented at 0° azimuth, and the competing signal was presented from surrounding speakers located at 45°, 135°, 225°, and 315° azimuths, at a level of 68 dBA. The OLSA test was scored to determine the SNR for the 80%-correct point for each individual.
The results of this speech recognition testing revealed a significant (p<.05) 1.4 dB SNR advantage when the OneMic directionality was activated (compared to omnidirectional processing). This mean benefit, although significant, does not reflect the benefit obtained by the average participant, as it was influenced by an outlier who unexplainably obtained a large advantage for omnidirectional. The median benefit was 1.8 dB, and if we view the individual data, we see that the 66% majority (10-15 participants) obtained a OneMic advantage of 1.7 dB or greater, with three individuals obtaining an advantage of 3 dB or higher. Recall that the sAI-DI of the CIC directionality was about ½ of that of the ITC, so a directional benefit of 2-3 dB compares reasonably well with the ITC speech recognition benefit discussed earlier.
In this research, it was of interest to determine if removing and re-inserting the CIC instruments had an effect on OneMic benefit, and therefore, the participants were tested twice during the same clinic visit. The results of the two sessions were highly correlated, and there were no significant differences in the mean advantage for the OneMic processing for the two different sessions—the second session was used for the statistical analysis mentioned above. If we examine individual performance for both test sessions, however, and select the best performance for each individual, we see that the mean maximum benefit for the OneMic system increases to 2.1 dB.
Because of the different test sites, speech material and different loudspeaker arrays for the competing message, it is not possible to do a direct comparison of the absolute speech recognition for CICs versus ITCs, or to compare these findings to our previous beamforming research with mini-BTE/RIC instruments. It is possible, however, that although the benefit is less with CIC directionality, which is expected, as the ITCs utilize the monaural directivity in addition to the binaural beamforming, the absolute performance with the CIC products (which is what matters for the patient), may be the same as for these other hearing aid styles.
Many patients prefer to use custom instruments rather than BTEs. In the past, one downside of the small custom products was that the patient would then be deprived of advanced directional beamforming available in larger hearing aid styles. As we have shown here, that is no longer a concern. Clinical speech testing at two different independent sites reveals that a significant improvement in speech recognition is available for advanced directional processing for both the ITC and CIC styles. For both of these styles, the binaural beamformer is automatic and adaptive, only activating when the listening situation indicates the need. In other situations, the omnidirectional processing will be present to maintain spatial awareness.
Based on the Siemens e2e wireless 3.0 system, the power consumption is so minimal that these benefits can be provided in the standard universal program. In this way, the overall practical benefit for the user is maximized.
Preves D. Directional microphone use in ITE hearing instruments. Hearing Review. 1997;4(7):21-27.
Rumoshovsky J. Directional microphones ITE aids. Hear Jour. 1977;30:48–50.
Ricketts T, Lindley G, Henry P. Impact of compression and hearing aid style on directional hearing aid benefit and performance. Ear Hear. 2001 Aug;22(4):348-61.
Pumford J, Seewald R, Scollie S, Jenstad L. Speech recognition with in-the-ear and behind-the-ear dual-microphone hearing aids. J Am Acad Audiol. 2000 11:23-35.
Kamkar-Parsi H, Fischer E, Aubreville M. New binaural strategies for enhanced hearing. Hearing Review. 2014;21(10):42-45. Available at: http://www.hearingreview.com/2014/10/new-binaural-strategies-enhanced-hearing
Froehlich M, Freels K, Powers T. Speech recognition benefit obtained from binaural beamforming hearing aids: Comparison to omnidirectional and individuals with normal hearing. May 2015. Available at:http://www.audiologyonline.com/articles/speech-recognition-benefit-obtained-from-14338
Powers TA, Fröhlich M. Clinical results with a new wireless binaural directional hearing system. Hearing Review. 2014;21(11):32-34.
Herbig R, Froehlich M. Binaural beamforming: The natural evolution. Hearing Review. 2015; 22(5):24.
Aubreville M, Petrausch S. Measuring directionality of modern hearing aids.Canadian Audiologist. 2015;2(3).
Aubreville M, Petrausch S. Binaural beamforming and directionality: Siemens technology holds clear performance advantage. Canadian Audiologist. 2015;2(4).
Hagerman B, Olofsson Å. A method to measure the effect of noise reduction algorithms using simultaneous speech and noise. Acta Acust United Ac. 2004:90(2):356-361.
Wu YH, Bentler RA. Using a signal cancellation technique to assess adaptive directivity of hearing aids, J Acoust Soc Am. 2007:122(1),496-511.
Aubreville M, Petrausch S. Directionality assessment of adaptive binaural beamforming with noise suppression in hearing aids. Acoustics, Speech and Signal Processing (ICASSP), Proceedings IEEE International Conference, 2015c: 211-215.
Nilsson M, Soli SD, Sullivan, J. A. development of the Hearing in Noise Test for the measurement of speech reception thresholds in quiet and in noise. J Acoust Soc Am. 1994;95(2):1085-1099.
Kollmeier B, Wesselkamp M. Development and evaluation of a German sentence test for objective and subjective speech intelligibility assessment. J Acousti Soc Am. 1997;102(4):2412-2421.