Original research paper
Speech reception threshold benefits in cochlear implant users with an adaptive beamformer in real life situations Gunnar Geißler 1, Iris Arweiler 2 , Phillipp Hehrmann 2, Thomas Lenarz 1, Volkmar Hamacher 2, Andreas Büchner 1 1
Medizinische Hochschule Hannover, Germany, 2Advanced Bionics, European Research Center, Hannover, Germany Objectives: To compare the Naida CI UltraZoom adaptive beamformer and T-Mic settings in a real life environment. Methods: Speech reception thresholds (SRTs) were measured in a moderately reverberant room, using the German Oldenburger sentence test. The speech signal was always presented from the front loudspeaker at 0° azimuth and fixed masking noise was presented either simultaneously from all eight loudspeakers around the subject at 0°, ±45°, ±90°, ±135°, and 180° azimuth or from five loudspeakers positioned at ±70°, ±135°, and 180° azimuth. In the third test setup, an additional roving noise was added to the six loudspeaker arrangement. Results: There was a significant difference in mean SRTs between the Naida CI T-Mic and UltraZoom in each of the three test setups. The largest improvements were seen in the six speaker roving and fixed noise conditions. Adding ClearVoice to the Naida CI T-Mic setting significantly improved the SRT in both fixed noise conditions, but not in the roving noise condition. In each setup, the lowest SRTs were obtained with the UltraZoom plus ClearVoice setting. Discussion: The degree of improvement was consistent with previous beamforming studies. In the most challenging listening situation, with noise from eight speakers and speech and noise presented coincidentally from the front, UltraZoom still provided a significant benefit. When a moving noise source was added, the improvement in SRT provided by UltraZoom was maintained. Conclusion: When tested in challenging and realistic noise environments, the Naida CI UltraZoom adaptive beamformer resulted in significantly lower mean SRTs than when the T-Mic alone was used. Keywords: Cochlear implant, Naida CI, UltraZoom, Beamformer, Beamforming, ClearVoice, SRT, T-Mic
Introduction The speech perception performance of cochlear implant (CI) users in quiet has continued to improve to a point where, for a significant number of recipients, open set speech understanding on sentence tests reaches 100% (Gifford et al., 2008). However, performance in noise is still an issue and when compared to normal hearing listeners; CI recipients still require significantly higher signal to noise ratios (SNRs) to achieve equivalent performance (Firszt et al., 2004; Schafer et al., 2012; Spahr and Dorman 2004). How to provide this improvement in performance in noise
Correspondence to: Iris Arweiler, Advanced Bionics GmbH, European Research Center, Feodor-Lynen-Straße 35, Hannover 30625, Germany. Email: [email protected]
© W. S. Maney & Son Ltd 2015 DOI 10.1179/1754762814Y.0000000088
has become a major focus for both CI manufacturers and clinicians. There has been much discussion and evidence gathered to show that the provision of binaural hearing can improve hearing in noise, by allowing CI users to utilize headshadow and binaural release from masking effects (Ching et al., 2007; Laske et al., 2009; Litovsky et al., 2006; Potts et al., 2009). However, this can only be achieved either with an expensive second CI or, for individuals with sufficient residual hearing, a contralateral hearing aid. An alternative and complementary approach is to focus on improving the SNR of the input before the signal enters the processing pathway. This can be done in a number of different ways, one of which is the use of directional microphones. These types of microphone
Cochlear Implants International
2015
VOL.
16
NO.
2
69
Geißler et al.
Speech reception threshold benefits in cochlear implant users
are not equally sensitive to sound arriving from all directions (omni-directional), but have reduced sensitivity to sound arriving from a specified location, usually attenuating sound from behind the listener which is often noise. The process by which the microphone directionality is created is called beamforming. The basic principles of all types of beamforming technology rely on time and therefore phase differences in the signals arriving at two or more spatially separated microphones. Modern beamformers, using multimicrophone arrays and digital processing, can shape the directional response characteristics of the microphone output to produce an unlimited variety of different sensitivity patterns, showing highest sensitivity only for signals arriving from a limited range of angles (Kates, 1993; Saunders and Kates, 1997; Valente et al., 1995). They can be used to produce a static pattern of directionality, where the point of maximum attenuation or null is fixed, or an adaptive one, where the null dynamically follows the direction of noise incidence (Kompis and Dillier, 2001). These multiple microphone beamformers have been incorporated successfully into hearing aids since the 1990s and two systematic reviews identified weak to moderate evidence of their effectiveness in improving speech perception in noise (Bentler, 2005; McCreery et al., 2012; Ricketts and Dhar, 1999; Ricketts and Henry, 2002). However, the first adaptive beamforming algorithms integrated into CI speech processors were not commercially introduced until 2005 (Spriet et al., 2007). In studies assessing the performance of this adaptive beamformer, for speech perception with multiple noise sources in a laboratory environment, improvements in the SNR were very large and ranged from 6.5 to 3.9 dB, depending on the specifications of the test setup used (Brockmeyer and Potts, 2011 (4.2 dB), Gifford and Revit, 2010 (3.9 dB), Hersbach et al., 2012 (5.3 dB), Spriet et al., 2007 (6.5 dB in speech-weighted noise)). Up until now, the Harmony speech processor, manufactured by Advanced bionics (Advanced Bionics LLC, Valencia, California, USA), did not have the ability to use beamforming technology. In this processor, directionality was introduced with the T-Mic, an omni-directional microphone placed at the opening of the external auditory canal. This placement allowed the user to take advantage of the ear’s natural directivity created by the frequency response characteristics of the pinna (Shaw, 1974) and this directional benefit has been demonstrated for users of in-the-ear hearing aids compared to behind-the-ear hearing aids (Pumford et al., 2000). In an eight loudspeaker setup, with speech presented from the front and noise presented simultaneously from the other seven speakers, the T-Mic setting yielded a 4.4 dB improvement in
70
Cochlear Implants International
2015
VOL.
16
NO.
2
SNR, similar to that provided by the Freedom adaptive beamformer (Gifford and Revit, 2010). In 2013, Advanced Bionics introduced the new Naida CI speech processor incorporating an adaptive beamformer called UltraZoom, which had previously been used in the Phonak Ambra hearing aid (Nyffeler 2010). Initial studies investigating the use of UltraZoom in users of the Advanced Bionics CI used a test setup where the CI processor was connected via a direct audio input to an Ambra hearing aid with UltraZoom. The sound was then processed by the adaptive directional microphone of the hearing aid, not the CI processor microphone. Using this set up, Hehrmann et al. (2012) tested UltraZoom in 12 unilateral users of the Harmony speech processor with speech presented from the front and noise simultaneously from five speakers positioned at ±70°, ±135°, and 180° azimuth around the listener and showed substantial improvements in SNR of 5.2 dB on average with the UltraZoom compared to an omni-directional microphone setting. The aims of the study reported here were to test the UltraZoom adaptive beamformer and the T-Mic, as implemented in the new Naida CI speech processor. The T-Mic setting represents the current standard for directional listening in Advanced Bionics speech processors; therefore, this was chosen as the baseline measure against which the new UltraZoom beamforming setting was tested. As well as comparing the Naida CI UltraZoom and T-Mic settings to each other and to the T-Mic setting on the Harmony speech processor, Naida CI UltraZoom and T-Mic with and without ClearVoice were also compared. ClearVoice is a single-microphone noise reduction algorithm implemented in the Advanced Bionics speech processors and represents an alternative approach to noise management. It acts on the signal after it has been band-pass filtered and is continuously applied to all inputs. It estimates the SNR separately in each frequency band, based on the modulation content and bands with low estimated SNRs are attenuated. A full description is provided in Büchner et al. (2010). The addition of ClearVoice to either UltraZoom or T-mic should provide an additional improvement in SNR. Three different test setups were used to represent a variety of real life environments. Testing was done in a moderately reverberant room and a diffuse noise field was chosen, as this best represents the real world. In the first setup, the speech signal was presented from the front loudspeaker in isolation while the noise was presented from the eight surrounding loudspeakers, with noise from all directions, including the speech from the front speaker, to resemble a cocktail-party or restaurant situation. In the second setup there were five surrounding loudspeakers, as used by Hehrmann et al. (2012), representing a lecture room
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Table 1 Subject demographics Subject ID
Age (years)
Duration of deafness
47 58 68 65 56 66 59 67 60 77
11 0 0 4 0 0 2 0 10 0
1 2 3 4 5 6 7 8 9 10
Etiology
Type of implant
Unknown Genetic Genetic Unknown Unknown Genetic Unknown Sudden hearing loss Unknown Unknown
AB_Clarion Cll AB_HiRes90K AB_Clarion Cll AB_Clarion Cll AB_HiRes90K AB_HiRes90K AB_HiRes90K AB_HiRes90K AB_Clarion Cll AB_HiRes90K
with the presenter in front and noise from people behind and to the side of the listener. A third setup used the same loudspeaker arrangement as in the second setup, but with an additional noise source randomly roving between the five loudspeakers. This resembles a situation where a conversation ends at one place and starts at another and was designed to test the adaptive nature of UltraZoom.
Test subjects Ten postlingually deafened CI users participated in the study. Their ages were between 47 and 77 years with a mean age of 62 years. All users were unilaterally implanted with Advanced Bionics HiRes90K or CII CIs. They had at least 6 months experience with the Harmony speech processor and the HiRes 120 processing strategy and a minimum of 3 months experience with ClearVoice. Their speech recognition scores with the HSM sentence test (Hochmair-Desoyer et al., 1997) in noise were 10% or better at an SNR of 10 dB, as measured at their last clinic appointment. The duration of deafness ranged between 0 and 11 years with a mean of 3 years. The mean age of implantation was 55 years. Demographic details of the test subjects are listed in Table 1.
Duration of implant use (years) 12 5 12 10 5 4 6 4 11 5
always presented from the front loudspeaker at 0° azimuth. For the first test setup, fixed stationary speech-shaped noise was presented simultaneously from all eight loudspeakers around the test subject at 0°, ±45°, ±90°, ±135°, and 180° azimuth. For the second test setup, speech-shaped noise was presented simultaneously from five loudspeakers positioned at ±70°, ±135°, and 180° azimuth, again with fixed noise and in the third test setup an additional roving noise was added to the five fixed noise loudspeakers placed at ±70°, ±135°, and 180° azimuth (Figs. 1–3). The test subject was seated in the middle of the loudspeaker array in an acoustically dampened room, with a T60 reverberation time of 0.6 seconds. The loudspeakers were positioned at a distance of 1.2 m from the test subject, who sat in the middle of the test array. For each of the test setups, levels were checked with an omni-directional microphone, placed at the location of the center of the listener’s head, with the listener absent.
Stimuli Speech perception was tested using sentences from the German sentence test ‘Oldenburger Satztest (OlSa)’ (Wagener et al., 1999). Speech reception thresholds
Procedures Subjects’ speech perception ability in noise was measured on three separate occasions, using the three different loudspeaker arrangements. During each session, subjects were tested using the new Naida CI speech processor with the omni-directional microphone, T-Mic and adaptive beamformer settings as well as a loaner Harmony speech processor on the T-Mic setting. The current program was loaded onto the loaner speech processor and testing was carried out acutely without any take home experience. Subjects then returned home after each session with their own, unadjusted, Harmony speech processor. Each subject attended three sessions, each lasting approximately 3 hours including breaks. Sessions were spread across between 3 days and 2 months. Within each session, the conditions tested were presented in a random order. The speech signal was
Figure 1 Loudspeaker setup 1: the speech signal was presented from front (0° azimuth) and the noise signal was presented from eight loudspeakers at 0°, ±45°, ±90°, ±135°, and 180° azimuth. Red speakers indicate the speech signal and black speakers indicate a noise signal.
Cochlear Implants International
2015
VOL.
16
NO.
2
71
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Figure 2 Loudspeaker setup 2: The speech signal was presented from front (0° azimuth) and the SSN noise signal was presented from five loudspeakers at ±70°, ±135°, and 180° azimuth. Red speakers indicate the speech signal and black speakers indicate a noise signal.
(SRTs) were measured with two lists of 20 sentences (test and re-test) per condition and averaged between those two lists. Each sentence consists of five words with a fixed syntactical structure (name–verb–number–adjective–object, e.g. Thomas gibt acht schwere Schuhe (Thomas gives eight heavy shoes)) spoken by a man. After each sentence presentation, the test subject had to repeat the words understood; tight scoring was used. In order to estimate the SRT, the level of the sentences was varied adaptively in a fixed level noise field of 65 dBA and the SNR measured at which 50% of the words were understood correctly. At the beginning of each session, two practice lists of 20 sentences were given to minimize learning effects during the test session and these lists were not used for any further testing. The test subject was instructed to look at the front loudspeaker and to hold their head upright during sentence presentation. The average
SRT was calculated for each condition and the same set of comparisons were made for each test setup. The noise signal used consisted of a stationary speech-shaped noise (SSN) presented at 65 dBA, created by repeatedly superimposing sequences of the OlSa sentences with the male speaker, and had the same average long-term spectrum as the sentences. The SSN was cut into uncorrelated noise signals and played simultaneously from all loudspeakers in each of the three test set ups. The noise was presented continuously during sentence presentation. For test setup three, the SSN was played at 60 dBA and an additional International Female Fluctuating Masker (IFFM) noise source was added (Holube et al., 2010). This noise was randomly roved across all five noise speakers at 1.5 second intervals. The total level of noise remained at 65 dBA.
Statistics Data from each loudspeaker test setup were treated as a separate sample. The averaged SRT for each condition, calculated as the SNR required to achieve 50% on the sentence test, was compared for primary and secondary outcome measures for each of the three test setups, as listed in Table 2. Results of the OlSa sentence test are typically normally distributed, so analysis was performed using a series of pairwise comparisons using a two-tailed, paired student t-test. In order to control for a type I error, a Bonferroni corrected P value of less than 0.01 was considered to be statistically significant. Mean values are given with confidence limits of one standard deviation. Power calculations based on the standard deviations of the samples showed that with 10 subjects the maximum effect size which could be detected reliably was 1.1 dB.
Results In the eight speaker test setup, with speech and noise presented from the front, the mean SRTs ± 1 SD for the Naida CI with T-Mic and with UltraZoom were −2.3 ± 0.9 and −6 ± 1.2 dB, respectively, and there Table 2 Primary and secondary outcome measures for comparison in each of the three test setups Objective
Condition 1
Condition 2
Primary
T-Mic of the Naida CI processor T-Mic of the Naida CI processor T-Mic of the Naida CI processor
UltraZoom on the Naida CI processor T-Mic of the Naida CI processor Omni-directional microphone of the Naida CI processor T-Mic of the Naida CI processor without ClearVoice UltraZoom on the Naida CI processor without ClearVoice
Secondary
Figure 3 Loudspeaker setup 3: The speech signal was presented from front (0° azimuth) and the SSN noise signal was presented from the five loudspeakers at ±70°, ±135°, and 180° azimuth. The IFFM signal was additionally presented from one out of the five loudspeakers in a pseudorandomized order. Red speakers indicate the speech signal and black speakers indicate a noise signal.
72
Cochlear Implants International
2015
VOL.
16
NO.
2
T-Mic of the Naida CI processor with ClearVoice UltraZoom on the Naida CI processor with ClearVoice
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Figure 4 Average speech reception thresholds (SRTs) for each of the conditions tested in the eight speaker fixed noise test setup. Error bars show 1 standard deviation. Starred brackets indicate a significant difference.
was a significant difference between them (P < 0.001) (Fig. 4). There was also a significant improvement in SRT to −3.7 ± 1.5 dB when ClearVoice was added to the Naida CI with T-Mic condition (P < 0.01). The lowest SRTs and therefore best SRTs were obtained in what could be considered the simplest setup with fixed, diffuse noise presented simultaneously from five speakers and speech from a sixth speaker directly in front of the subject (Fig. 5). Mean SRTs for the Naida CI with T-Mic and with UltraZoom were −2.4 ± 1.1 and −7.9 ± 1.4 dB, respectively, and there was a significant difference between them (P < 0.001). The addition of ClearVoice to the Naida CI T-Mic setting significantly improved the SRT to −3.6 ± 1.5 dB (P < 0.01) and its addition to the Naida CI UltraZoom significantly improved the SRT to −8.7 ± 1.7 dB (P < 0.01). When a roving noise source was added to the six speaker test setup, mean SRTs for the Naida with TMic and with UltraZoom were 0.5 ± 1.6 and −5.1 ± 2.1 dB, respectively (Fig. 6). This difference was
significant with P < 0.001. There were no other significant differences between means for this test setup. There was no significant difference between the Naida CI T-Mic and Naida CI Omni-directional setting or the Naida CI T-Mic and Harmony T-Mic in any of the three set ups. Overall, the lowest SRTs in each setup were obtained with the Naida CI UltraZoom plus ClearVoice setting.
Discussion The results of the study showed that the UltraZoom beamformer significantly improved the SRTs achieved compared to the T-Mic alone in each of the test setups used. The degree of improvement was consistent with previous studies using a CI connected to the Ambra hearing aid and the adaptive beamformers used in alternative CI designs (Brockmeyer and Potts, 2011, Gifford and Revit, 2010, Hehrmann et al., 2012, Hersbach et al., 2012, Spriet et al., 2007.). An important scenario tested in this study is the situation where a noise source is coming from the same
Figure 5 Average speech reception thresholds (SRTs) for each of the conditions tested in the six speaker fixed noise test setup. Error bars show 1 standard deviation. Starred brackets indicate a significant difference.
Cochlear Implants International
2015
VOL.
16
NO.
2
73
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Figure 6 Average speech reception thresholds (SRTs) for each of the conditions tested in the six speaker roving noise test setup. Error bars show 1 standard deviation. Starred brackets indicate a significant difference.
direction as the speech. This is one of the most challenging situations for any beamformer as they mainly rely on timing and phase differences between the speech and noise signals and therefore work best when speech and noise are spatially separated (Brockmeyer and Potts, 2011). We can see that the introduction of noise coincidentally with the speech and the additional speakers at ±45° azimuth in the eight speaker arrangement reduced the SRT advantage of the UltraZoom over the other settings by around 2 dB. In contrast, the T-Mic setting maintains a consistent SRT advantage of −2.3 dB. However, when the roving noise source was added, despite an overall worsening in SRT for all the conditions, the adaptive beamformer was the most robust, maintaining the same levels of SRT gain over the Naida CI T-Mic setting as in the six speaker fixed noise scenario. The design of the test environments was chosen to be particularly challenging to better represent a realworld noisy environment, rather than to maximize the possible measurable benefit in a laboratory setting. All tests were conducted in a room with a moderate average T60 reverberation time of 0.6 seconds, higher than the T60 of 0.5 seconds reported in Spriet et al. (2007) and the 0.4 seconds reported by Chung et al. (2012). A reverberation time of this length could be considered to be similar to the maximum value for a classroom environment in line with the American national standard on classroom acoustics (American National Standards Institute S12.60-2002 (2002)). In reverberant rooms the performance of beamformers has been shown to degrade, as both the target and interfering noise become more spatially diffuse and consequentially more difficult to separate based on their directionality (Greenberg and Zurek, 1992; Kompis and Dillier, 2001). The impact of reverberation is likely to be a factor as to why CI recipients report smaller subjective benefits from beamformers in real life than can be expected from the gain measured
74
Cochlear Implants International
2015
VOL.
16
NO.
2
in controlled laboratory conditions (Spriet et al., 2007). Well-matched and calibrated microphones are also essential, as even slight differences in phase or sensitivity can cause dramatic directivity losses, especially in the 0.5–2 kHz range (Peterson et al., 1990). While the study aimed to create a realistic noisy environment for testing, because no subjective measures were used and subjects had no take home experience with the new processor, their ability to transfer the advantages of beamforming recorded in the laboratory into the real world remains unknown. Other studies have also shown that beamforming can be used effectively in CIs, but comparison of our results to these studies is difficult due to the differing loudspeaker arrangements used (Hersbach et al., 2012; Spriet et al., 2007). A key aspect of this study was the use of speakers positioned in the frontal hemisphere, i.e. forward of ±90° azimuth and, in the case of the eight speaker arrangement, speech and noise presented coincidentally from the front speaker. Concern had previously been expressed by researchers in the hearing aid field that laboratory testing did not evaluate the listeners performance in a way that reflected real world listening (Brockmeyer and Potts, 2011). To address this, the R-SAPCE test system was designed to replicate a restaurant environment (Compton-Conley et al., 2004). It uses a diffuse noise field, in an eight speaker setup with speech and noise from the front, similar to the test setup used here. Brockmeyer and Potts (2011) and Gifford and Revit (2010) used R-SPACE to test the BEAM adaptive beamformer in the Nucleus Freedom speech processor. The 3.6 dB improvement in SRT in comparison to the T-Mic setting shown here is slightly smaller than the 4.2 and 3.9 dB improvements in SRT when compared to recipients standard CI program, reported by them. However, a non-reverberant test environment was used for both R-SPACE studies. When the UltraZoom was compared to the Naida CI omni-
Geißler et al.
directional setting, a 4.4 dB improvement was seen. It should also be remembered that speech recognition tasks may be harder when R-SPACE noise is used compared to the continuous noise type used here (Valente et al., 2006). The T-Mic setting performed the same on the Naida CI and Harmony processors and there was no significant difference in SRT between them. No significant advantage could be shown for the T-Mic setting on the Naida CI over its omni-directional setting. This is in contrast to data previously reported by Gifford and Revit (2010), where a significant gain of 4.4 dB in SRT for the T-Mic over the omni-directional setting was shown in their R-SPACE set up, the same as the improvement over the omni-directional setting obtained from UltraZoom in this study. However, there were some key differences between the test environments; the speakers in the Gifford and Revit (2010) study were positioned much closer to the subject (60 cm) and a sound-treated room was used. This will have impacted the results as directional benefits are at their maximum with small speaker to listener distances and minimal reverberation (Hawkins and Yacullo, 1984; Ricketts and Henry, 2002). When the ClearVoice noise reduction algorithm was applied to the T-Mic and UltraZoom settings, a small additive benefit was observed. This was significant for the T-Mic with ClearVoice setting in both six and eight speaker fixed noise setups. However, no advantage was seen when the roving noise source was added. When added to the UltraZoom, a significant benefit was only observed for the six speaker fixed noise setup, but the effect size was below the 1.1 dB difference which could be reliably detected by statistical analysis based on the power calculation. It is known that ClearVoice provides larger benefits overall at higher SNRs (Brendel et al., 2012) and the very low SNRs required to measure the SRT in the UltraZoom conditions will reduce the amount of benefit that could be expected from it. There was no benefit from adding ClearVoice to either UltraZoom or T-Mic in the third test setup, which made use of a mixture of stationary noise and fluctuating IFFM noise. ClearVoice acts on the signal after it has been bandpass filtered and works on the assumption that the speech envelope is modulated, while the noise envelope is stationary. In order to allow the UltraZoom to adapt, the fluctuating noise had to be the dominant noise source and this may have made it harder for ClearVoice to separate the speech and stationary noise.
Conclusions When tested in a variety of challenging and realistic noise environments in a moderately reverberant room, the UltraZoom adaptive beamformer
Speech reception threshold benefits in cochlear implant users
implemented on the Naida CI speech processor resulted in significantly lower mean SRTs than when the T-Mic alone setting was used. ClearVoice provided an additional benefit over the T-Mic setting when a fixed noise test setup was used. When a moving noise source was added to the six speaker test setup, the improvement in SRT provided by UltraZoom over the T-Mic setting was maintained. When the most challenging listening situation was tested, with noise from eight speakers and speech and noise presented coincidentally from the front, UltraZoom still provided a significant benefit when compared to the baseline T-Mic setting. In each of the loudspeaker arrangements tested, the lowest SRTs were obtained with the UltraZoom plus ClearVoice setting.
Disclaimer statements Contributors GG is the main researcher, IA and PH contributed to research and writing support, TL is a paper reviewer; VH and AB are paper reviewers and contributed to research design support. Funding None. Conflicts of interest Some of the authors are employees of Advanced Bionics and provided writing assistance. Ethics approval Ethics approval was obtained for this study from the MHH Ethikkommission OE 9515 on 12 April 2013.
References Bentler R.A. 2005. Effectiveness of directional microphones and noise reduction schemes in hearing aids: a systematic review of the evidence. Journal of the American Academy of Audiology, 16(7): 473–484. Brendel M., Rottmann T., Frohne-Büchner C., Büchner A., Lenarz T. 2012. Benefit of the Noise Reduction Algorithm ClearVoice for experienced cochlear implant users. Presented at the 12th International Conference on Cochlear Implants and Other Implantable Auditory Technologies; May 3–5; Baltimore, MD, USA. Brockmeyer A.M., Potts L.G. 2011. Evaluation of different signal processing options in unilateral and bilateral cochlear freedom implant recipients using R-Space background noise. Journal of the American Academy of Audiology, 22(2): 65–80. Büchner A., Brendel M., Saalfeld H., Litvak L., Frohne-Büchner C., Lenarz T. 2010. Results of a pilot study with a signal enhancement algorithm for HiRes120 cochlear implant users. Otology and Neurotology, 31(9): 1386–1390. Ching T.Y., van Wanrooy E., Dillon H. 2007. Binaural-bimodal fitting or bilateral implantation for managing severe to profound deafness: a review. Trends in Amplification, 11(3): 161–192. Chung K., Nelson L., Teske M. 2012. Noise reduction technologies implemented in head-worn preprocessors for improving cochlear implant performance in reverberant noise fields. Hearing Research, 291(1–2): 41–51. Compton-Conley C.L., Neuman A.C., Killion M.C., Levitt H. 2004. Performance of directional microphones for hearing aids: real-world versus simulation. Journal of the American Academy of Audiology, 15(6): 440–455. Firszt J.B., Holden L.K., Skinner M.W., Tobey E.A., Peterson A., Gaggl W., et al. 2004. Recognition of speech presented at
Cochlear Implants International
2015
VOL.
16
NO.
2
75
Geißler et al.
Speech reception threshold benefits in cochlear implant users
soft to loud levels by adult cochlear implant recipients of three cochlear implant systems. Ear and Hearing, 25(4): 375–387. Gifford R.H., Revit L.J. 2010. Speech perception for adult cochlear implant recipients in a realistic background noise: effectiveness of preprocessing strategies and external options for improving speech recognition in noise. Journal of the American Academy of Audiology, 21(7): 441–451. Gifford R.H., Shallop J.K., Peterson A.M. 2008. Speech recognition materials and ceiling effects: considerations for cochlear implant programs. Audiology and Neuro-otology, 13(3): 193–205. Greenberg J.E., Zurek P.M. 1992. Evaluation of an adaptive beamforming method for hearing aids. Journal of Acoustical Society of America, 91(3): 1662–1676. Hawkins D.B., Yacullo W.S. 1984. Signal-to-noise ratio advantage of binaural hearing aids and directional microphones under different levels of reverberation. Journal of Speech and Hearing Disorders, 49(3): 278–286. Hehrmann P., Fredelake S., Hamacher V., Dyballa K.H., Büchner A. 2012. Improved speech intelligibility with cochlear implants using state-of-the-art noise reduction algorithms. ITG Report 236, Speech Communication, Braunschweig. Hersbach A.A., Arora K., Mauger S.J., Dawson P.W. 2012. Combining directional microphone and single-channel noise reduction algorithms: a clinical evaluation in difficult listening conditions with cochlear implant users. Ear and Hearing, 33(4): e13–e23. Hochmair-Desoyer I., Schulz E., Moser L., Schmidt M. 1997. The HSM sentence test as a tool for evaluating the speech understanding in noise of cochlear implant users. American Journal of Otology, 18(6 Suppl): S83. Holube I., Fredelake S., Vlaming M., Kollmeier B. 2010. Development and analysis of an International Speech Test Signal (ISTS). International Journal of Audiology, 49(12): 891–903. Kates J.M. 1993. Superdirective arrays for hearing aids. Journal of the Acoustical Society of America, 94(4): 1930–1933. Kompis M., Dillier N. 2001. Performance of an adaptive beamforming noise reduction scheme for hearing aid applications. II. Experimental verification of the predictions. Journal of the Acoustical Society of America, 109(3): 1134–1143. Laske R.D., Veraguth D., Dillier N., Binkert A., Holzmann D., Huber A.M. 2009. Subjective and objective results after bilateral cochlear implantation in adults. Otology and Neurotology, 30: 313–318. Litovsky R., Parkinson A., Arcaroli J., Sammeth C. 2006. Simultaneous bilateral cochlear implantation in adults: a multicenter clinical study. Ear and Hearing, 27(6): 714–731. McCreery R.W., Venediktov R.A., Coleman J.J., Leech H.M. 2012. An evidence-based systematic review of directional microphones and digital noise reduction hearing aids in school-age
76
Cochlear Implants International
2015
VOL.
16
NO.
2
children with hearing loss. American Journal of Audiology, 21(2): 295–312. Nyffeler M. 2010. StereoZoom, improvements with directional microphones. Field study news. Phonak AG, Switzerland. Peterson P.M., Wie S.M., Rabinowitz W.M., Zurek P.M. 1990. Robustness of an adaptive beamforming method for hearing aids. Acta Otolaryngologica, Supplementum, 469: 85–90. Potts L.G., Skinner M.W., Litovsky R.A., Strube M.J., Kuk F. 2009. Recognition and localization of speech by adult cochlear implant recipients wearing a digital hearing aid in the nonimplanted ear (bimodal hearing). Journal of the American Academy of Audiology, 20(6): 353–373. Pumford J.M., Seewald R.C., Scollie S.D., Jenstad L.M. 2000. Speech recognition with in-the-ear and behind-the-ear dualmicrophone hearing instruments. Journal of the American Academy of Audiology, 11(1): 23–35. Ricketts T., Dhar S. 1999. Comparison of performance across three directional hearing aids. Journal of the American Academy of Audiology, 10(4): 180–189. Ricketts T., Henry P. 2002. Evaluation of an adaptive, directionalmicrophone hearing aid. International Journal of Audiology, 41(2): 100–112. Saunders G.H., Kates J.M. 1997. Speech intelligibility enhancement using hearing-aid array processing. Journal of the Acoustical Society of America, 102(3): 1827–1837. Schafer E.C., Pogue J., Milrany T. 2012. List equivalency of the AzBio sentence test in noise for listeners with normal-hearing sensitivity or cochlear implants. Journal of the American Academy of Audiology, 23(7): 501–509. Shaw E.A. 1974. Transformation of sound pressure level from the free field to the eardrum in the horizontal plane. Journal of the Acoustical Society of America, 56(6): 1848–1861. Spahr A.J., Dorman M.F. 2004. Performance of subjects fit with the Advanced Bionics CII and Nucleus 3G cochlear implant devices. Archives of Otolaryngology – Head and Neck Surgery, 130(5): 624–628. Spriet A., Van Deun L., Eftaxiadis K., Laneau J., Moonen M., van Dijk B., et al. 2007. Speech understanding in background noise with the two-microphone adaptive beamformer BEAM in the Nucleus Freedom Cochlear Implant System. Ear and Hearing, 28(1): 62–72. Valente M., Fabry D.A., Potts L.G. 1995. Recognition of speech in noise with hearing aids using dual microphones. Journal of the American Academy of Audiology, 6(6): 440–449. Valente M., Mispagel K.M., Tchorz J., Fabry D. 2006. Effect of type of noise and loudspeaker array on the performance of omnidirectional and directional microphones. Journal of the American Academy of Audiology, 17(6): 398–412. Wagener K., Brand T., Kollmeier B. 1999. Entwicklung und Evaluation eines Satztests für die deutsche Sprache. Teil III: Evaluation des Oldenburger Satztests. Zeitschrift für Audiologie, 38(3): 86–95.
Copyright of Cochlear Implants International: An Interdisciplinary Journal is the property of Maney Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Speech reception threshold benefits in cochlear implant users with an adaptive beamformer in real life situations Gunnar Geißler 1, Iris Arweiler 2 , Phillipp Hehrmann 2, Thomas Lenarz 1, Volkmar Hamacher 2, Andreas Büchner 1 1
Medizinische Hochschule Hannover, Germany, 2Advanced Bionics, European Research Center, Hannover, Germany Objectives: To compare the Naida CI UltraZoom adaptive beamformer and T-Mic settings in a real life environment. Methods: Speech reception thresholds (SRTs) were measured in a moderately reverberant room, using the German Oldenburger sentence test. The speech signal was always presented from the front loudspeaker at 0° azimuth and fixed masking noise was presented either simultaneously from all eight loudspeakers around the subject at 0°, ±45°, ±90°, ±135°, and 180° azimuth or from five loudspeakers positioned at ±70°, ±135°, and 180° azimuth. In the third test setup, an additional roving noise was added to the six loudspeaker arrangement. Results: There was a significant difference in mean SRTs between the Naida CI T-Mic and UltraZoom in each of the three test setups. The largest improvements were seen in the six speaker roving and fixed noise conditions. Adding ClearVoice to the Naida CI T-Mic setting significantly improved the SRT in both fixed noise conditions, but not in the roving noise condition. In each setup, the lowest SRTs were obtained with the UltraZoom plus ClearVoice setting. Discussion: The degree of improvement was consistent with previous beamforming studies. In the most challenging listening situation, with noise from eight speakers and speech and noise presented coincidentally from the front, UltraZoom still provided a significant benefit. When a moving noise source was added, the improvement in SRT provided by UltraZoom was maintained. Conclusion: When tested in challenging and realistic noise environments, the Naida CI UltraZoom adaptive beamformer resulted in significantly lower mean SRTs than when the T-Mic alone was used. Keywords: Cochlear implant, Naida CI, UltraZoom, Beamformer, Beamforming, ClearVoice, SRT, T-Mic
Introduction The speech perception performance of cochlear implant (CI) users in quiet has continued to improve to a point where, for a significant number of recipients, open set speech understanding on sentence tests reaches 100% (Gifford et al., 2008). However, performance in noise is still an issue and when compared to normal hearing listeners; CI recipients still require significantly higher signal to noise ratios (SNRs) to achieve equivalent performance (Firszt et al., 2004; Schafer et al., 2012; Spahr and Dorman 2004). How to provide this improvement in performance in noise
Correspondence to: Iris Arweiler, Advanced Bionics GmbH, European Research Center, Feodor-Lynen-Straße 35, Hannover 30625, Germany. Email: [email protected]
© W. S. Maney & Son Ltd 2015 DOI 10.1179/1754762814Y.0000000088
has become a major focus for both CI manufacturers and clinicians. There has been much discussion and evidence gathered to show that the provision of binaural hearing can improve hearing in noise, by allowing CI users to utilize headshadow and binaural release from masking effects (Ching et al., 2007; Laske et al., 2009; Litovsky et al., 2006; Potts et al., 2009). However, this can only be achieved either with an expensive second CI or, for individuals with sufficient residual hearing, a contralateral hearing aid. An alternative and complementary approach is to focus on improving the SNR of the input before the signal enters the processing pathway. This can be done in a number of different ways, one of which is the use of directional microphones. These types of microphone
Cochlear Implants International
2015
VOL.
16
NO.
2
69
Geißler et al.
Speech reception threshold benefits in cochlear implant users
are not equally sensitive to sound arriving from all directions (omni-directional), but have reduced sensitivity to sound arriving from a specified location, usually attenuating sound from behind the listener which is often noise. The process by which the microphone directionality is created is called beamforming. The basic principles of all types of beamforming technology rely on time and therefore phase differences in the signals arriving at two or more spatially separated microphones. Modern beamformers, using multimicrophone arrays and digital processing, can shape the directional response characteristics of the microphone output to produce an unlimited variety of different sensitivity patterns, showing highest sensitivity only for signals arriving from a limited range of angles (Kates, 1993; Saunders and Kates, 1997; Valente et al., 1995). They can be used to produce a static pattern of directionality, where the point of maximum attenuation or null is fixed, or an adaptive one, where the null dynamically follows the direction of noise incidence (Kompis and Dillier, 2001). These multiple microphone beamformers have been incorporated successfully into hearing aids since the 1990s and two systematic reviews identified weak to moderate evidence of their effectiveness in improving speech perception in noise (Bentler, 2005; McCreery et al., 2012; Ricketts and Dhar, 1999; Ricketts and Henry, 2002). However, the first adaptive beamforming algorithms integrated into CI speech processors were not commercially introduced until 2005 (Spriet et al., 2007). In studies assessing the performance of this adaptive beamformer, for speech perception with multiple noise sources in a laboratory environment, improvements in the SNR were very large and ranged from 6.5 to 3.9 dB, depending on the specifications of the test setup used (Brockmeyer and Potts, 2011 (4.2 dB), Gifford and Revit, 2010 (3.9 dB), Hersbach et al., 2012 (5.3 dB), Spriet et al., 2007 (6.5 dB in speech-weighted noise)). Up until now, the Harmony speech processor, manufactured by Advanced bionics (Advanced Bionics LLC, Valencia, California, USA), did not have the ability to use beamforming technology. In this processor, directionality was introduced with the T-Mic, an omni-directional microphone placed at the opening of the external auditory canal. This placement allowed the user to take advantage of the ear’s natural directivity created by the frequency response characteristics of the pinna (Shaw, 1974) and this directional benefit has been demonstrated for users of in-the-ear hearing aids compared to behind-the-ear hearing aids (Pumford et al., 2000). In an eight loudspeaker setup, with speech presented from the front and noise presented simultaneously from the other seven speakers, the T-Mic setting yielded a 4.4 dB improvement in
70
Cochlear Implants International
2015
VOL.
16
NO.
2
SNR, similar to that provided by the Freedom adaptive beamformer (Gifford and Revit, 2010). In 2013, Advanced Bionics introduced the new Naida CI speech processor incorporating an adaptive beamformer called UltraZoom, which had previously been used in the Phonak Ambra hearing aid (Nyffeler 2010). Initial studies investigating the use of UltraZoom in users of the Advanced Bionics CI used a test setup where the CI processor was connected via a direct audio input to an Ambra hearing aid with UltraZoom. The sound was then processed by the adaptive directional microphone of the hearing aid, not the CI processor microphone. Using this set up, Hehrmann et al. (2012) tested UltraZoom in 12 unilateral users of the Harmony speech processor with speech presented from the front and noise simultaneously from five speakers positioned at ±70°, ±135°, and 180° azimuth around the listener and showed substantial improvements in SNR of 5.2 dB on average with the UltraZoom compared to an omni-directional microphone setting. The aims of the study reported here were to test the UltraZoom adaptive beamformer and the T-Mic, as implemented in the new Naida CI speech processor. The T-Mic setting represents the current standard for directional listening in Advanced Bionics speech processors; therefore, this was chosen as the baseline measure against which the new UltraZoom beamforming setting was tested. As well as comparing the Naida CI UltraZoom and T-Mic settings to each other and to the T-Mic setting on the Harmony speech processor, Naida CI UltraZoom and T-Mic with and without ClearVoice were also compared. ClearVoice is a single-microphone noise reduction algorithm implemented in the Advanced Bionics speech processors and represents an alternative approach to noise management. It acts on the signal after it has been band-pass filtered and is continuously applied to all inputs. It estimates the SNR separately in each frequency band, based on the modulation content and bands with low estimated SNRs are attenuated. A full description is provided in Büchner et al. (2010). The addition of ClearVoice to either UltraZoom or T-mic should provide an additional improvement in SNR. Three different test setups were used to represent a variety of real life environments. Testing was done in a moderately reverberant room and a diffuse noise field was chosen, as this best represents the real world. In the first setup, the speech signal was presented from the front loudspeaker in isolation while the noise was presented from the eight surrounding loudspeakers, with noise from all directions, including the speech from the front speaker, to resemble a cocktail-party or restaurant situation. In the second setup there were five surrounding loudspeakers, as used by Hehrmann et al. (2012), representing a lecture room
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Table 1 Subject demographics Subject ID
Age (years)
Duration of deafness
47 58 68 65 56 66 59 67 60 77
11 0 0 4 0 0 2 0 10 0
1 2 3 4 5 6 7 8 9 10
Etiology
Type of implant
Unknown Genetic Genetic Unknown Unknown Genetic Unknown Sudden hearing loss Unknown Unknown
AB_Clarion Cll AB_HiRes90K AB_Clarion Cll AB_Clarion Cll AB_HiRes90K AB_HiRes90K AB_HiRes90K AB_HiRes90K AB_Clarion Cll AB_HiRes90K
with the presenter in front and noise from people behind and to the side of the listener. A third setup used the same loudspeaker arrangement as in the second setup, but with an additional noise source randomly roving between the five loudspeakers. This resembles a situation where a conversation ends at one place and starts at another and was designed to test the adaptive nature of UltraZoom.
Test subjects Ten postlingually deafened CI users participated in the study. Their ages were between 47 and 77 years with a mean age of 62 years. All users were unilaterally implanted with Advanced Bionics HiRes90K or CII CIs. They had at least 6 months experience with the Harmony speech processor and the HiRes 120 processing strategy and a minimum of 3 months experience with ClearVoice. Their speech recognition scores with the HSM sentence test (Hochmair-Desoyer et al., 1997) in noise were 10% or better at an SNR of 10 dB, as measured at their last clinic appointment. The duration of deafness ranged between 0 and 11 years with a mean of 3 years. The mean age of implantation was 55 years. Demographic details of the test subjects are listed in Table 1.
Duration of implant use (years) 12 5 12 10 5 4 6 4 11 5
always presented from the front loudspeaker at 0° azimuth. For the first test setup, fixed stationary speech-shaped noise was presented simultaneously from all eight loudspeakers around the test subject at 0°, ±45°, ±90°, ±135°, and 180° azimuth. For the second test setup, speech-shaped noise was presented simultaneously from five loudspeakers positioned at ±70°, ±135°, and 180° azimuth, again with fixed noise and in the third test setup an additional roving noise was added to the five fixed noise loudspeakers placed at ±70°, ±135°, and 180° azimuth (Figs. 1–3). The test subject was seated in the middle of the loudspeaker array in an acoustically dampened room, with a T60 reverberation time of 0.6 seconds. The loudspeakers were positioned at a distance of 1.2 m from the test subject, who sat in the middle of the test array. For each of the test setups, levels were checked with an omni-directional microphone, placed at the location of the center of the listener’s head, with the listener absent.
Stimuli Speech perception was tested using sentences from the German sentence test ‘Oldenburger Satztest (OlSa)’ (Wagener et al., 1999). Speech reception thresholds
Procedures Subjects’ speech perception ability in noise was measured on three separate occasions, using the three different loudspeaker arrangements. During each session, subjects were tested using the new Naida CI speech processor with the omni-directional microphone, T-Mic and adaptive beamformer settings as well as a loaner Harmony speech processor on the T-Mic setting. The current program was loaded onto the loaner speech processor and testing was carried out acutely without any take home experience. Subjects then returned home after each session with their own, unadjusted, Harmony speech processor. Each subject attended three sessions, each lasting approximately 3 hours including breaks. Sessions were spread across between 3 days and 2 months. Within each session, the conditions tested were presented in a random order. The speech signal was
Figure 1 Loudspeaker setup 1: the speech signal was presented from front (0° azimuth) and the noise signal was presented from eight loudspeakers at 0°, ±45°, ±90°, ±135°, and 180° azimuth. Red speakers indicate the speech signal and black speakers indicate a noise signal.
Cochlear Implants International
2015
VOL.
16
NO.
2
71
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Figure 2 Loudspeaker setup 2: The speech signal was presented from front (0° azimuth) and the SSN noise signal was presented from five loudspeakers at ±70°, ±135°, and 180° azimuth. Red speakers indicate the speech signal and black speakers indicate a noise signal.
(SRTs) were measured with two lists of 20 sentences (test and re-test) per condition and averaged between those two lists. Each sentence consists of five words with a fixed syntactical structure (name–verb–number–adjective–object, e.g. Thomas gibt acht schwere Schuhe (Thomas gives eight heavy shoes)) spoken by a man. After each sentence presentation, the test subject had to repeat the words understood; tight scoring was used. In order to estimate the SRT, the level of the sentences was varied adaptively in a fixed level noise field of 65 dBA and the SNR measured at which 50% of the words were understood correctly. At the beginning of each session, two practice lists of 20 sentences were given to minimize learning effects during the test session and these lists were not used for any further testing. The test subject was instructed to look at the front loudspeaker and to hold their head upright during sentence presentation. The average
SRT was calculated for each condition and the same set of comparisons were made for each test setup. The noise signal used consisted of a stationary speech-shaped noise (SSN) presented at 65 dBA, created by repeatedly superimposing sequences of the OlSa sentences with the male speaker, and had the same average long-term spectrum as the sentences. The SSN was cut into uncorrelated noise signals and played simultaneously from all loudspeakers in each of the three test set ups. The noise was presented continuously during sentence presentation. For test setup three, the SSN was played at 60 dBA and an additional International Female Fluctuating Masker (IFFM) noise source was added (Holube et al., 2010). This noise was randomly roved across all five noise speakers at 1.5 second intervals. The total level of noise remained at 65 dBA.
Statistics Data from each loudspeaker test setup were treated as a separate sample. The averaged SRT for each condition, calculated as the SNR required to achieve 50% on the sentence test, was compared for primary and secondary outcome measures for each of the three test setups, as listed in Table 2. Results of the OlSa sentence test are typically normally distributed, so analysis was performed using a series of pairwise comparisons using a two-tailed, paired student t-test. In order to control for a type I error, a Bonferroni corrected P value of less than 0.01 was considered to be statistically significant. Mean values are given with confidence limits of one standard deviation. Power calculations based on the standard deviations of the samples showed that with 10 subjects the maximum effect size which could be detected reliably was 1.1 dB.
Results In the eight speaker test setup, with speech and noise presented from the front, the mean SRTs ± 1 SD for the Naida CI with T-Mic and with UltraZoom were −2.3 ± 0.9 and −6 ± 1.2 dB, respectively, and there Table 2 Primary and secondary outcome measures for comparison in each of the three test setups Objective
Condition 1
Condition 2
Primary
T-Mic of the Naida CI processor T-Mic of the Naida CI processor T-Mic of the Naida CI processor
UltraZoom on the Naida CI processor T-Mic of the Naida CI processor Omni-directional microphone of the Naida CI processor T-Mic of the Naida CI processor without ClearVoice UltraZoom on the Naida CI processor without ClearVoice
Secondary
Figure 3 Loudspeaker setup 3: The speech signal was presented from front (0° azimuth) and the SSN noise signal was presented from the five loudspeakers at ±70°, ±135°, and 180° azimuth. The IFFM signal was additionally presented from one out of the five loudspeakers in a pseudorandomized order. Red speakers indicate the speech signal and black speakers indicate a noise signal.
72
Cochlear Implants International
2015
VOL.
16
NO.
2
T-Mic of the Naida CI processor with ClearVoice UltraZoom on the Naida CI processor with ClearVoice
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Figure 4 Average speech reception thresholds (SRTs) for each of the conditions tested in the eight speaker fixed noise test setup. Error bars show 1 standard deviation. Starred brackets indicate a significant difference.
was a significant difference between them (P < 0.001) (Fig. 4). There was also a significant improvement in SRT to −3.7 ± 1.5 dB when ClearVoice was added to the Naida CI with T-Mic condition (P < 0.01). The lowest SRTs and therefore best SRTs were obtained in what could be considered the simplest setup with fixed, diffuse noise presented simultaneously from five speakers and speech from a sixth speaker directly in front of the subject (Fig. 5). Mean SRTs for the Naida CI with T-Mic and with UltraZoom were −2.4 ± 1.1 and −7.9 ± 1.4 dB, respectively, and there was a significant difference between them (P < 0.001). The addition of ClearVoice to the Naida CI T-Mic setting significantly improved the SRT to −3.6 ± 1.5 dB (P < 0.01) and its addition to the Naida CI UltraZoom significantly improved the SRT to −8.7 ± 1.7 dB (P < 0.01). When a roving noise source was added to the six speaker test setup, mean SRTs for the Naida with TMic and with UltraZoom were 0.5 ± 1.6 and −5.1 ± 2.1 dB, respectively (Fig. 6). This difference was
significant with P < 0.001. There were no other significant differences between means for this test setup. There was no significant difference between the Naida CI T-Mic and Naida CI Omni-directional setting or the Naida CI T-Mic and Harmony T-Mic in any of the three set ups. Overall, the lowest SRTs in each setup were obtained with the Naida CI UltraZoom plus ClearVoice setting.
Discussion The results of the study showed that the UltraZoom beamformer significantly improved the SRTs achieved compared to the T-Mic alone in each of the test setups used. The degree of improvement was consistent with previous studies using a CI connected to the Ambra hearing aid and the adaptive beamformers used in alternative CI designs (Brockmeyer and Potts, 2011, Gifford and Revit, 2010, Hehrmann et al., 2012, Hersbach et al., 2012, Spriet et al., 2007.). An important scenario tested in this study is the situation where a noise source is coming from the same
Figure 5 Average speech reception thresholds (SRTs) for each of the conditions tested in the six speaker fixed noise test setup. Error bars show 1 standard deviation. Starred brackets indicate a significant difference.
Cochlear Implants International
2015
VOL.
16
NO.
2
73
Geißler et al.
Speech reception threshold benefits in cochlear implant users
Figure 6 Average speech reception thresholds (SRTs) for each of the conditions tested in the six speaker roving noise test setup. Error bars show 1 standard deviation. Starred brackets indicate a significant difference.
direction as the speech. This is one of the most challenging situations for any beamformer as they mainly rely on timing and phase differences between the speech and noise signals and therefore work best when speech and noise are spatially separated (Brockmeyer and Potts, 2011). We can see that the introduction of noise coincidentally with the speech and the additional speakers at ±45° azimuth in the eight speaker arrangement reduced the SRT advantage of the UltraZoom over the other settings by around 2 dB. In contrast, the T-Mic setting maintains a consistent SRT advantage of −2.3 dB. However, when the roving noise source was added, despite an overall worsening in SRT for all the conditions, the adaptive beamformer was the most robust, maintaining the same levels of SRT gain over the Naida CI T-Mic setting as in the six speaker fixed noise scenario. The design of the test environments was chosen to be particularly challenging to better represent a realworld noisy environment, rather than to maximize the possible measurable benefit in a laboratory setting. All tests were conducted in a room with a moderate average T60 reverberation time of 0.6 seconds, higher than the T60 of 0.5 seconds reported in Spriet et al. (2007) and the 0.4 seconds reported by Chung et al. (2012). A reverberation time of this length could be considered to be similar to the maximum value for a classroom environment in line with the American national standard on classroom acoustics (American National Standards Institute S12.60-2002 (2002)). In reverberant rooms the performance of beamformers has been shown to degrade, as both the target and interfering noise become more spatially diffuse and consequentially more difficult to separate based on their directionality (Greenberg and Zurek, 1992; Kompis and Dillier, 2001). The impact of reverberation is likely to be a factor as to why CI recipients report smaller subjective benefits from beamformers in real life than can be expected from the gain measured
74
Cochlear Implants International
2015
VOL.
16
NO.
2
in controlled laboratory conditions (Spriet et al., 2007). Well-matched and calibrated microphones are also essential, as even slight differences in phase or sensitivity can cause dramatic directivity losses, especially in the 0.5–2 kHz range (Peterson et al., 1990). While the study aimed to create a realistic noisy environment for testing, because no subjective measures were used and subjects had no take home experience with the new processor, their ability to transfer the advantages of beamforming recorded in the laboratory into the real world remains unknown. Other studies have also shown that beamforming can be used effectively in CIs, but comparison of our results to these studies is difficult due to the differing loudspeaker arrangements used (Hersbach et al., 2012; Spriet et al., 2007). A key aspect of this study was the use of speakers positioned in the frontal hemisphere, i.e. forward of ±90° azimuth and, in the case of the eight speaker arrangement, speech and noise presented coincidentally from the front speaker. Concern had previously been expressed by researchers in the hearing aid field that laboratory testing did not evaluate the listeners performance in a way that reflected real world listening (Brockmeyer and Potts, 2011). To address this, the R-SAPCE test system was designed to replicate a restaurant environment (Compton-Conley et al., 2004). It uses a diffuse noise field, in an eight speaker setup with speech and noise from the front, similar to the test setup used here. Brockmeyer and Potts (2011) and Gifford and Revit (2010) used R-SPACE to test the BEAM adaptive beamformer in the Nucleus Freedom speech processor. The 3.6 dB improvement in SRT in comparison to the T-Mic setting shown here is slightly smaller than the 4.2 and 3.9 dB improvements in SRT when compared to recipients standard CI program, reported by them. However, a non-reverberant test environment was used for both R-SPACE studies. When the UltraZoom was compared to the Naida CI omni-
Geißler et al.
directional setting, a 4.4 dB improvement was seen. It should also be remembered that speech recognition tasks may be harder when R-SPACE noise is used compared to the continuous noise type used here (Valente et al., 2006). The T-Mic setting performed the same on the Naida CI and Harmony processors and there was no significant difference in SRT between them. No significant advantage could be shown for the T-Mic setting on the Naida CI over its omni-directional setting. This is in contrast to data previously reported by Gifford and Revit (2010), where a significant gain of 4.4 dB in SRT for the T-Mic over the omni-directional setting was shown in their R-SPACE set up, the same as the improvement over the omni-directional setting obtained from UltraZoom in this study. However, there were some key differences between the test environments; the speakers in the Gifford and Revit (2010) study were positioned much closer to the subject (60 cm) and a sound-treated room was used. This will have impacted the results as directional benefits are at their maximum with small speaker to listener distances and minimal reverberation (Hawkins and Yacullo, 1984; Ricketts and Henry, 2002). When the ClearVoice noise reduction algorithm was applied to the T-Mic and UltraZoom settings, a small additive benefit was observed. This was significant for the T-Mic with ClearVoice setting in both six and eight speaker fixed noise setups. However, no advantage was seen when the roving noise source was added. When added to the UltraZoom, a significant benefit was only observed for the six speaker fixed noise setup, but the effect size was below the 1.1 dB difference which could be reliably detected by statistical analysis based on the power calculation. It is known that ClearVoice provides larger benefits overall at higher SNRs (Brendel et al., 2012) and the very low SNRs required to measure the SRT in the UltraZoom conditions will reduce the amount of benefit that could be expected from it. There was no benefit from adding ClearVoice to either UltraZoom or T-Mic in the third test setup, which made use of a mixture of stationary noise and fluctuating IFFM noise. ClearVoice acts on the signal after it has been bandpass filtered and works on the assumption that the speech envelope is modulated, while the noise envelope is stationary. In order to allow the UltraZoom to adapt, the fluctuating noise had to be the dominant noise source and this may have made it harder for ClearVoice to separate the speech and stationary noise.
Conclusions When tested in a variety of challenging and realistic noise environments in a moderately reverberant room, the UltraZoom adaptive beamformer
Speech reception threshold benefits in cochlear implant users
implemented on the Naida CI speech processor resulted in significantly lower mean SRTs than when the T-Mic alone setting was used. ClearVoice provided an additional benefit over the T-Mic setting when a fixed noise test setup was used. When a moving noise source was added to the six speaker test setup, the improvement in SRT provided by UltraZoom over the T-Mic setting was maintained. When the most challenging listening situation was tested, with noise from eight speakers and speech and noise presented coincidentally from the front, UltraZoom still provided a significant benefit when compared to the baseline T-Mic setting. In each of the loudspeaker arrangements tested, the lowest SRTs were obtained with the UltraZoom plus ClearVoice setting.
Disclaimer statements Contributors GG is the main researcher, IA and PH contributed to research and writing support, TL is a paper reviewer; VH and AB are paper reviewers and contributed to research design support. Funding None. Conflicts of interest Some of the authors are employees of Advanced Bionics and provided writing assistance. Ethics approval Ethics approval was obtained for this study from the MHH Ethikkommission OE 9515 on 12 April 2013.
References Bentler R.A. 2005. Effectiveness of directional microphones and noise reduction schemes in hearing aids: a systematic review of the evidence. Journal of the American Academy of Audiology, 16(7): 473–484. Brendel M., Rottmann T., Frohne-Büchner C., Büchner A., Lenarz T. 2012. Benefit of the Noise Reduction Algorithm ClearVoice for experienced cochlear implant users. Presented at the 12th International Conference on Cochlear Implants and Other Implantable Auditory Technologies; May 3–5; Baltimore, MD, USA. Brockmeyer A.M., Potts L.G. 2011. Evaluation of different signal processing options in unilateral and bilateral cochlear freedom implant recipients using R-Space background noise. Journal of the American Academy of Audiology, 22(2): 65–80. Büchner A., Brendel M., Saalfeld H., Litvak L., Frohne-Büchner C., Lenarz T. 2010. Results of a pilot study with a signal enhancement algorithm for HiRes120 cochlear implant users. Otology and Neurotology, 31(9): 1386–1390. Ching T.Y., van Wanrooy E., Dillon H. 2007. Binaural-bimodal fitting or bilateral implantation for managing severe to profound deafness: a review. Trends in Amplification, 11(3): 161–192. Chung K., Nelson L., Teske M. 2012. Noise reduction technologies implemented in head-worn preprocessors for improving cochlear implant performance in reverberant noise fields. Hearing Research, 291(1–2): 41–51. Compton-Conley C.L., Neuman A.C., Killion M.C., Levitt H. 2004. Performance of directional microphones for hearing aids: real-world versus simulation. Journal of the American Academy of Audiology, 15(6): 440–455. Firszt J.B., Holden L.K., Skinner M.W., Tobey E.A., Peterson A., Gaggl W., et al. 2004. Recognition of speech presented at
Cochlear Implants International
2015
VOL.
16
NO.
2
75
Geißler et al.
Speech reception threshold benefits in cochlear implant users
soft to loud levels by adult cochlear implant recipients of three cochlear implant systems. Ear and Hearing, 25(4): 375–387. Gifford R.H., Revit L.J. 2010. Speech perception for adult cochlear implant recipients in a realistic background noise: effectiveness of preprocessing strategies and external options for improving speech recognition in noise. Journal of the American Academy of Audiology, 21(7): 441–451. Gifford R.H., Shallop J.K., Peterson A.M. 2008. Speech recognition materials and ceiling effects: considerations for cochlear implant programs. Audiology and Neuro-otology, 13(3): 193–205. Greenberg J.E., Zurek P.M. 1992. Evaluation of an adaptive beamforming method for hearing aids. Journal of Acoustical Society of America, 91(3): 1662–1676. Hawkins D.B., Yacullo W.S. 1984. Signal-to-noise ratio advantage of binaural hearing aids and directional microphones under different levels of reverberation. Journal of Speech and Hearing Disorders, 49(3): 278–286. Hehrmann P., Fredelake S., Hamacher V., Dyballa K.H., Büchner A. 2012. Improved speech intelligibility with cochlear implants using state-of-the-art noise reduction algorithms. ITG Report 236, Speech Communication, Braunschweig. Hersbach A.A., Arora K., Mauger S.J., Dawson P.W. 2012. Combining directional microphone and single-channel noise reduction algorithms: a clinical evaluation in difficult listening conditions with cochlear implant users. Ear and Hearing, 33(4): e13–e23. Hochmair-Desoyer I., Schulz E., Moser L., Schmidt M. 1997. The HSM sentence test as a tool for evaluating the speech understanding in noise of cochlear implant users. American Journal of Otology, 18(6 Suppl): S83. Holube I., Fredelake S., Vlaming M., Kollmeier B. 2010. Development and analysis of an International Speech Test Signal (ISTS). International Journal of Audiology, 49(12): 891–903. Kates J.M. 1993. Superdirective arrays for hearing aids. Journal of the Acoustical Society of America, 94(4): 1930–1933. Kompis M., Dillier N. 2001. Performance of an adaptive beamforming noise reduction scheme for hearing aid applications. II. Experimental verification of the predictions. Journal of the Acoustical Society of America, 109(3): 1134–1143. Laske R.D., Veraguth D., Dillier N., Binkert A., Holzmann D., Huber A.M. 2009. Subjective and objective results after bilateral cochlear implantation in adults. Otology and Neurotology, 30: 313–318. Litovsky R., Parkinson A., Arcaroli J., Sammeth C. 2006. Simultaneous bilateral cochlear implantation in adults: a multicenter clinical study. Ear and Hearing, 27(6): 714–731. McCreery R.W., Venediktov R.A., Coleman J.J., Leech H.M. 2012. An evidence-based systematic review of directional microphones and digital noise reduction hearing aids in school-age
76
Cochlear Implants International
2015
VOL.
16
NO.
2
children with hearing loss. American Journal of Audiology, 21(2): 295–312. Nyffeler M. 2010. StereoZoom, improvements with directional microphones. Field study news. Phonak AG, Switzerland. Peterson P.M., Wie S.M., Rabinowitz W.M., Zurek P.M. 1990. Robustness of an adaptive beamforming method for hearing aids. Acta Otolaryngologica, Supplementum, 469: 85–90. Potts L.G., Skinner M.W., Litovsky R.A., Strube M.J., Kuk F. 2009. Recognition and localization of speech by adult cochlear implant recipients wearing a digital hearing aid in the nonimplanted ear (bimodal hearing). Journal of the American Academy of Audiology, 20(6): 353–373. Pumford J.M., Seewald R.C., Scollie S.D., Jenstad L.M. 2000. Speech recognition with in-the-ear and behind-the-ear dualmicrophone hearing instruments. Journal of the American Academy of Audiology, 11(1): 23–35. Ricketts T., Dhar S. 1999. Comparison of performance across three directional hearing aids. Journal of the American Academy of Audiology, 10(4): 180–189. Ricketts T., Henry P. 2002. Evaluation of an adaptive, directionalmicrophone hearing aid. International Journal of Audiology, 41(2): 100–112. Saunders G.H., Kates J.M. 1997. Speech intelligibility enhancement using hearing-aid array processing. Journal of the Acoustical Society of America, 102(3): 1827–1837. Schafer E.C., Pogue J., Milrany T. 2012. List equivalency of the AzBio sentence test in noise for listeners with normal-hearing sensitivity or cochlear implants. Journal of the American Academy of Audiology, 23(7): 501–509. Shaw E.A. 1974. Transformation of sound pressure level from the free field to the eardrum in the horizontal plane. Journal of the Acoustical Society of America, 56(6): 1848–1861. Spahr A.J., Dorman M.F. 2004. Performance of subjects fit with the Advanced Bionics CII and Nucleus 3G cochlear implant devices. Archives of Otolaryngology – Head and Neck Surgery, 130(5): 624–628. Spriet A., Van Deun L., Eftaxiadis K., Laneau J., Moonen M., van Dijk B., et al. 2007. Speech understanding in background noise with the two-microphone adaptive beamformer BEAM in the Nucleus Freedom Cochlear Implant System. Ear and Hearing, 28(1): 62–72. Valente M., Fabry D.A., Potts L.G. 1995. Recognition of speech in noise with hearing aids using dual microphones. Journal of the American Academy of Audiology, 6(6): 440–449. Valente M., Mispagel K.M., Tchorz J., Fabry D. 2006. Effect of type of noise and loudspeaker array on the performance of omnidirectional and directional microphones. Journal of the American Academy of Audiology, 17(6): 398–412. Wagener K., Brand T., Kollmeier B. 1999. Entwicklung und Evaluation eines Satztests für die deutsche Sprache. Teil III: Evaluation des Oldenburger Satztests. Zeitschrift für Audiologie, 38(3): 86–95.
Copyright of Cochlear Implants International: An Interdisciplinary Journal is the property of Maney Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
