Speech Recognition in Complex Listening Situations of Bimodal and Single-Sided Deaf Cochlear-Implant Recipients

Speech Recognition in Complex Listening Situations of Bimodal and Single-Sided Deaf Cochlear-Implant Recipients Fabian Eberling1,2, Max Blümer4, Mark Praetorius4, Michael Schulte1,2, Jan Heeren1,2, Kirsten C. Wagener1,2, Thomas Brand2,3 1Hörzentrum Oldenburg gGmbH, Oldenburg 2Cluster of Excellence Hearing4All, Oldenburg 3Department of Medical Physics and Acoustics, Carl von Ossietzky University Oldenburg, Oldenburg 4Department of Otorhinolaryngology, University Medical Center Hamburg-Eppendorf, Hamburg Current Affiliations: Fabian Eberling: Audiological Acoustics, Department of Otolaryngology, University Hospital Frankfurt, Goethe University Frankfurt Max Blümer: Department of Otorhinolaryngology, Head and Neck Surgery, University Hospital Essen, University Duisburg-Essen Abstract Cochlear implant (CI) evaluations typically use monosyllabic word recognition (MWR) in quiet, which does not reflect the complex listening environments CI users encounter, such as diffuse background noises, multiple talkers, and divided attention (DA) scenarios. This study investigated speech recognition (SR) in CI recipients using the Oldenburg Sentence Test (OLSA) and a multitalker paradigm, targeting factors beyond MWR. 8 CI users with normal hearing in the contralateral ear (Single-Sided Deafness, SSD) and 10 with contralateral hearing aid (bimodal, BIM) participated. MWR in quiet was measured using the Freiburg Monosyllabic Test (contralateral ear blocked). SR in noise was assessed using OLSA with different spatial scenarios and masker types. Spatial release from masking (SRM) was extracted from speech reception thresholds (SRT). Selective attention (SA) benefits were evaluated using the Concurrent OLSA (CCOLSA) best aided in Singleand Dual-Task conditions. Hearing-related cognitive performance in diffuse noise was assessed using adaptive CCOLSA. MWR did not differ significantly between groups. SSD patients exhibited significantly better SRTs than BIM patients. Spatial separation improved SR for both groups. SSD patients performed better in competing speech than stationary noise; BIM patients did not. Both groups showed a better-ear effect. SSD patients demonstrated significantly higher hearing-related cognitive performance and benefited from SA. CCOLSA is feasible for SSD and BIM patients. Comparable MWR does not imply similar SR in more ecologically valid situations. The results suggest that the hearing loss in the acoustic ear of BIM patients limits SA abilities and increases cognitive load, even when best aided. Keywords: Cochlear Implant; Bimodal Hearing; Single-Sided-Deafness; Speech recognition; Selective Attention 1 1 Introduction Restoring speech recognition (SR) is the main objective when it comes to hearing system provision for people with impaired hearing. In Germany, cochlear implant (CI) surgery guidelines as defined in "Weißbuch Cochlea-Implantat (CI)-Versorgung” (DGHNO-KHC, 2021) demand for a SR test in quiet for the indication and postoperative determination of the provision success. Therefore, monosyllabic word recognition (MWR) is typically evaluated using the Freiburg Monosyllabic Test (Freiburger Einsilbertest, FBE, Hahlbrock, 1953). The test is usually conducted in quiet with an occluded contralateral ear. While it is an adequate test to assess basic SR abilities, essential aspects of SR in real-life situations, such as diffuse background noises, conversational turn-taking, varying voices or divided attention (DA) are not taken into account, but often highlighted by patients as demanding situations (Wagener et al., 2008; Pang et al., 2019). Therefore, when investigating SR abilities in everyday life, measurement methods that are more ecologically valid than the FBE and that take into account a broader range of the aforementioned factors should be used. Using a memory test in which target words had to be repeated, McCoy et al. (2005) were able to show that normal-hearing and hearing-impaired participants could achieve comparable results in a test of speech intelligibility. However, people with hearing loss had to put in more effort as they repeated significantly fewer words that were not part of the target words. Overall, they were able to memorize fewer words. This raises the question, whether CI patients could reach comparable results in SR tasks but show performance differences in complex listening situations. Heeren et al. (2022) presented the Concurrent Oldenburg Sentence Test (CCOLSA), which yields the possibility of measuring attention abilities and cognitive abilities in situations with turn-taking talkers. CCOLSA has already been successfully administered to patients with HA provision, and its feasibility will now be assessed for patients with CI provision. Furthermore, the differences in SR abilities between two patient groups relevant in Germany will be explored. CI candidacy criteria vary significantly across countries, particularly with regard to single-sided deafness (SSD). Although the frequency of CI implantation for SSD patients is increasing, not all countries have established guidelines for these patients (Van De Heyning et al., 2022). Germany, however, has specific indications for SSD candidates (DGHNO-KHC /AWMF, 2020). Notably, a substantial proportion of adult CI users (74.5%) have unilateral CI provision, with 54% also using a hearing aid (HA) on the contralateral side (Alfakhri et al., 2024). Therefore, this study focusses on patients with unilateral CI provision, encompassing those with single-sided deafness and those with bimodal hearing (BIM). Both groups combine electrical and acoustical hearing, but with different hearing abilities on the acoustically hearing side. While SSD patients are normally hearing on the non-implanted side, BIM patients have impaired hearing compensated by a HA. In 2008, Van de Heyning et al. conducted one of the pioneering studies on CI surgery for patients 2 with unilateral deafness. Interestingly, the primary aim was not necessarily to enhance SR but to alleviate tinnitus symptoms. Nonetheless, alongside significant relief from tinnitus, a noticeable improvement in SR was observed during the study (Vermeire & Van De Heyning, 2009). Since then, a lot of studies investigated the effect of unilateral CI provision on SR in patients with unilateral deafness (e.g. Buechner et al., 2010; Arndt et al., 2011; Hoth et al., 2016), because untreated unilateral deafness usually leads to difficulties in SR, especially in noisy environments, and negatively affects the localization of sounds (e.g. Cabral Junior et al., 2016; Cañete et al., 2019; Snapp and Ausili, 2020). In these cases, the provision of a CI can partially restore binaural hearing, resulting in enhanced speech recognition, improved sound localization abilities (e.g. Zeitler et al., 2015; Arndt et al., 2017; Buss et al., 2018; Litovsky et al., 2019; Thompson et al., 2022) as well as subjectively reported better hearing (e.g. Rösli et al., 2015; Dillon et al., 2017; Prejban et al., 2018; Häußler et al., 2020). Despite the observed improvements in hearing outcomes following cochlear implantation, the auditory abilities of SSD and BIM patients remain constrained due to limited access to binaural cues, such as interaural level differences (ILDs) and interaural time differences (ITDs). ITDs, which depend on precise temporal encoding, are poorly represented in CI stimulation strategies, particularly at the high pulse rates commonly employed in speech coding strategies (Laback et al., 2004, 2015). In contrast, ILDs, based on amplitude differences, are more reliably encoded and more easily perceived by CI users (Grantham et al., 2007; Seeber & Fastl, 2008). Consequently, ILDs are processed more effectively than ITDs due to inherent limitations in CI systems, which can only partially resolve the temporal fine structure of input signals while predominantly preserving envelope cues (Van Hoesel, 2004; Ausili et al., 2020). For BIM patients the access becomes even more difficult due to the signal processing of the HA (Udesen et al., 2013), but the access to ILDs and ITDs is relevant for SR in noise as well as the localization and separation of sound sources (Avan et al., 2015). Even though both groups benefit from binaural hearing, they also tend to Better-Ear-Listening (BE), which means that the subject’s listening performance is predominantly determined by the better hearing ear rather than by the impression of both ears together, especially in noisy environments with separated target and noise sources (Williges et al., 2019). However, the measurement conditions used by Williges et al. (2019), like stationary maskers presented from fixed positions, are not necessarily representative for situations faced in everyday life. In everyday life, we must deal with complex listening situations, consisting not only of stationary noise sources, but also containing competing talkers, fluctuating background sounds and music (Wagener et al., 2008). Besides different signalto-noise ratios (SNR), factors like attention and cognitive load play a crucial role in these socalled cocktail-party situations (Bronkhorst, 2000). After checking the feasibility of the CCOLSA in CI patients, this measurement method will be used to investigate whether and how differences in SR between SSD and BIM patients can be observed for complex listening situations. The investigations also focus on the crucial factors leading to possible performance differences 3 (masking through stationary noises or interfering speech as well as attention abilities). Stickney et al. (2004) found, through comparison of SR of normally hearing participants and bilateral CI users, that one interfering talker had a larger influence on bilateral CI users than on normally hearing participants. Both subject groups showed worse results for one interfering talker than for speech-shaped noise. When changing the interfering talker’s voice from the same male as the target talker to another male or female, the normally hearing participants showed an improvement in speech intelligibility while the different voices had no influence on the performance of the bilateral CI participants. Koelewijn et al. (2014) found comparable results for participants with hearing impairment. Using one female target talker and one male interfering talker, the participants showed worse speech intelligibility than for speech-shaped noise. The aim of the present study was the comparison of the SR performance of SSD and BIM patients, who show comparable MWR on the CI side, in complex listening situations measured using CCOLSA. Furthermore, the validity of the outcome of conventional measurement procedures used daily in clinical context, like the FBE or the Oldenburg Sentence Test (Oldenburger Satztest, OLSA), will be analyzed. 4 2 Methods 2.1 Participants Eighteen participants (nine of each sex) with a mean age of 60.2 years were assessed. The youngest participant was 24 years old and the oldest 81 years. No limitations regarding age, gender, duration of CI or HA provision, implant, speech processor or HA were made. Each participant had to have a minimum of 60% MWR in quiet while listening only via CI (contralateral ear blocked), on the last annual check-up appointment. For the separation into SSDand BIMparticipants, the pure-tone-average (PTA; frequencies 500, 1k, 2k, 4k Hz) of the acoustically hearing side was added as criterion to the existing HA provision. Also, each participant had to be able to perform speech recognition tests in noise. A PTA < 30 dB HL was set as limit for normal acoustical hearing in the contralateral ear. All participants with normal acoustical hearing were assigned to the SSD-group. Participants with a PTA  30 dB HL and existing HA provision in the contralateral ear were assigned to the BIM-group. Contralateral hearing thresholds of all participants are depicted in Figure 1, including the arithmetic mean of the two participant groups. Figure 1: The figure shows the hearing thresholds of the acoustical hearing side of the SSD (dashed) and BIM (dotted) groups. Also, the arithmetic mean of the SSD (green) and BIM (blue) group is shown. One of the SSD-participants showed a previously unknown hearing loss, which led to a PTA of 33.3 dB HL. Since it is only a mild hearing loss, the participant participated in the study as part of the SSD-group. Hence, the highest PTA in the SSD-group was 33.3 dB HL and the lowest PTA for the BIM-group was 35.8 dB HL. In the end, 8 participants were assigned to the SSD-group and 10 participants to the BIM-group. The exact demographic data as well as the provision 5 information are listed in Table 1. Each participant was recruited through the university medical center Hamburg-Eppendorf. Participation was on a voluntary basis and was not remunerated. The experiment was approved by the ethics committee of the University of Hamburg. Table 1: Demographic data and etiology of hearing loss of each participant ID Age Etiology Sex Deafness (years) CI ear Implant CI Processor CI Usage (years) Group PTA (dB HL) VP 01 69 Sudden hearing loss F ca. 20 years L Mi1250 Rondo 3 1,1 BiM 56,7 VP 02 81 Chronic otitis media M min. 10 years R CI622 CP1150 1,6 BiM 64,2 VP 03 24 unknown F N/S R CI512 CP1000 12,8 BiM 66,7 VP 04 27 Chronic otitis media M 3-4 years R CI622 CP1150 0,7 SSD 4,2 VP 05 56 Sudden hearing loss F N/S L Mi1250 Sonnet 2 0,9 BiM 54,2 VP 06 64 Sudden hearing loss F N/S R Mi1250 Sonnet 2 1,7 SSD 22,9 VP 07 32 Sudden hearing loss F N/S L CI622 CP1150 0,5 SSD 17,9 VP 08 77 Otosclerosis F N/S R Mi1250 Rondo 3 0,9 BiM 65,4 VP 09 43 unknown M 1,5 years L Mi1250 Sonnet 2 0,6 BiM 80,4 VP 10 60 Morbus Menière F N/S L Mi1250 Rondo 2 5,5 SSD 32,9 VP 11 70 Revision stapes surgery F N/S R CI622 CP1150 0,9 BiM 35,4 VP 12 80 Vestibular schwannoma M N/S R Mi1250 Rondo 3 1 SSD 25,4 VP 13 62 Sudden hearing loss M N/S L CI622 CP1000 3,9 SSD 0,0 VP 14 65 Morbus Menière M N/S R Mi1250 Sonnet 2 1,9 SSD 6,6 VP 15 80 Presbyacusis M N/S L Mi1250 Rondo 3 1,2 BiM 44,2 VP 16 81 unknown M N/S L CI622 CP1150 0,5 BiM 70,4 VP 17 59 Sudden hearing loss F N/S L Mi1250 Rondo 2 8,8 SSD 1,3 VP 18 53 Sudden hearing loss M N/S L CI622 CP1000 2,2 BiM 57,9 F = female; M = male; N/S = not specified; CI = Cochlear Implant; L = left; R = right; SSD = Single-SidedDeafness; BIM = bimodal; PTA = pure tone average 2.2 Measurement Setup All measurements were conducted in the audiology laboratory of the university medical center Hamburg-Eppendorf. The testing room (5 m x 4 m x 3,5 m) was optimized for acoustical measurements, using sound absorbers on the ceiling, molton curtains on the walls and carpet flooring. The participants were seated on a chair in the middle of the room. In 1 m distance, 16 Genelec 8030B (GENELEC, Finland) loudspeakers were placed in equidistant distances around the participants position, starting in front of the participant. The loudspeaker output was controlled using TASCAR-scenes (Grimm et al., 2015) run on a Dell (Dell Technologies Inc., USA) desktop computer with an Ubuntu operating system. Selection as well as activation/deactivation of the TASCAR-scenes was controlled using OSC-commands in Matlab 2023b (The MathWorks Inc., USA) run on a Dell (Dell Technologies Inc., USA) desktop computer with Windows 10 (Microsoft Corporation, USA) as operating system, which had a direct cable connection to the Ubuntu computer. Using an RME ADI-8 converter (Audio AG, Germany), the windows computer was connected to the loudspeakers. For the loudspeaker calibration, a 2260 sound analyzer (Brüel & Kjær, Denmark) was used, calibrated with a type 4231 calibrator (Brüel & Kjær, Denmark) at 94 dB SPL. All measurements were performed using a linear scale for the levels. The sound analyzer was placed at the participants position at the height of the loudspeakers. At first, the loudspeaker levels were equalized to 75 dB SPL with maximum deviations of 0,5 dB, using white noise as calibration signal. Subsequently the output level of all signals has been calibrated. Pure tone audiometry was conducted using a custom MATLAB script in MATLAB 2023b (The MathWorks Inc., USA), which presented pure tones through model DT770 (Beyerdynamic, 6 Germany) headphones. The measured frequencies ranged from 125 Hz to 8 kHz. The calibration was done using a 2610 measurement amplifier (Brüel & Kjær, Denmark) and an artificial ear type 4153 with coupler (Brüel & Kjær, Denmark). Sine tones were presented at 60 dB SPL as calibration signal. For the determination of dB HL values, it was assumed that correct thresholds could be approximated by taking into account the differences in the frequency response between the DT770 and the HDA200. Therefore, the RETSPL values of the HDA200 were used, with an additional correction applied to compensate for the frequency response differences between the HDA200 and the DT770 at each audiometric frequency measured. 2.6 General Measurement Procedure All measurements and respective conditions are listed in Table 2. Before starting any measurement, the ear canal of the acoustical hearing ear was checked using otoscopy, to ensure optimal measuring conditions. Afterwards, pure tone audiometry of the acoustical hearing ear Table 2: Measurement Conditions Measurement Conditions Note Pure tone audiometry Headphone measurement Hearing threshold, PTA FBE S0N0 Monosyllabic word recognition OLSA (Cafeteria Noise) S0Ndiff SRT in diffuse cafeteria noise OLSA (olnoise) S0N0/S0Nac/S0NCI SRT with stationary masker OLSA (FIT) S0N0/S0Nac/S0NCI SRT with female interfering talker CCOLSA Adaptive Overlap time at 50% correct, CCO CCOLSA DT Word scoring with DA CCOLSA ST Word scoring with SA PTA = Pure Tone Average; FBE = Freiburg Monosyllabic Test; SRT = Speech Reception Threshold; FIT = Female Interfering Talker; CCO = CCOLSAcosts; DA = Divided Attention; SA = Selective Attention; DA = Dual Task; ST = Single Task and the FBE with CI only (acoustical hearing ear blocked) were conducted to determine the parameters used for group allocation. As a next step, the individual SRTs for all three CCOLSA talkers (see Section 2.5) were determined in diffuse cafeteria noise. Afterwards the adaptive CCOLSA (see Section 2.5.3) was measured to determine the individual overlap time. The speech reception thresholds (SRTs) as well as the overlap time are needed for the remaining CCOLSA conditions. All remaining measurements (OLSA with olnoise/FIT, CCOLSA ST/DT) were conducted in randomized order. Participants of the SSD group overall had to perform two more measurements than the BIM group: The CCOLSA ST and DT condition were conducted unaided (SSD) and with HA provision (SSDaided) of the acoustical normal hearing ear. The HA provision enables the investigation of the influence of a HAs signal processing on the combination of electrical and acoustical hearing. A random HA was selected for the SSDaided condition and was fitted to the individual hearing threshold using the prescription fitting formula provided by the manufacturer and using closed domes as acoustic coupling. The choice were the HAs Pure Charge&Go 5 Nx 7 (Signia, Germany), Opn S1 miniRITE (Oticon, Denmark) and Audéo M70-R (Phonak, Switzerland). Features like e.g. automatic scene detection or noise reduction were deactivated if possible or set to the lowest level otherwise. Microphones were set to omnidirectional. The gain values based on the feedback suppression were discarded. Neither the HAs of the BIM group nor the CIs of both groups were fitted to match certain references. Instead, the most preferred settings of the participants were used. 2.3 Freiburg Monosyllabic Test Monaural MWR at 65 dB SPL in quiet, with CI only (contralateral ear occluded), was assessed using the FBE in free field. The measurement setup is shown in Figure 2(a). Speech was presented from the front and a combination of ear plugs and earmuffs was used to exclude the acoustical hearing ear in the contralateral side. The participant’s task was to repeat the presented words. For cases in which the participant couldn’t understand the presented words correctly, guessing was allowed. For each participant one test list, including 20 German monosyllabic words, was measured. Following the recommendation of Winkler and Holube (2014), only the lists 6, 7 and 16 were measured in pseudorandomized order. Measurements were controlled using an application run on a Dell (Dell Technologies Inc., USA) desktop computer, designed using the Matlab 2023b (The MathWorks Inc., USA) app designer. (a) Freiburg Monosyllabic Test (b) OLSA S0N0, S0N90ac, S0N90CI (c) OLSA diffuse (d) CCOLSA Figure 2: The figure schematically illustrates the speaker arrangement. The speakers active during the respective measurement condition are highlighted in red. a) Freiburg monosyllabic word test. b) S0N0, S0N90ac and S0N90CI conditions of OLSA. c) OLSA in diffuse background noise, d) CCOLSA measurement: due to the diffuse background noise, all speakers are active. Additionally, the positions of the two male talkers and the female talker are indicated. 14 Figure 6: SRM for the SSD (green) and BIM (blue) group, with maskers olnoise (left) and FIT (right), depending on the noise source position (N90ac, N90CI). The boxes extend to the interquartile range, with the median represented by a horizontal line. The whiskers extend to a maximum of 1.5 times the length of the box and outliers are marked with ’x’. Statistical significance is indicated by asterisks (*p<.05). 3.5 CCOLSA Word Scoring The word scoring results from the CCOLSA ST and DT measurements are shown in Figure 7. The differences in median word scoring results between the ST and DT conditions were in the SSD group 30.0%, in the SSDaided condition 37.5% and only 6.7% in the BIM group. The impact of HA provision is evident in the reduced word scoring outcomes for SSD compared to SSDaided. This was observed in both the ST (median difference: 4.2%) and the DT (median difference: 11.7%). For each dataset, normal distribution could be assumed (SW-Test, p  0.051). Since the results of SSD and SSDaided are paired samples, but not paired with the BIM results, two separate ANOVAs were conducted to avoid mixing paired and unpaired samples within a single ANOVA. To investigate the differences between the SSD and BIM group, a two-way mixed ANOVA (Type III) with between-subject factor group (SSD, BIM) and within-subject factor task (ST, DT) was performed. No significant main effects of either group or task were found, but a significant interaction between the factors group and task was found (F(1, 16) = 11.07, p = 0.004), indicating a different dependence of the word scoring within each group on the task. To further examine this interaction two one-way ANOVAs were conducted to investigate the influence of both factors separately. The effect of task within each group appeared to be significant for the SSD group (F(1, 7) = 64.3, p < 0.001) but not for the BIM group (F(1, 9) = 1.05, p = 0.332). For the factor group, a significant effect could only be observed in the ST (F(1, 16) = 8.4, p = 0.01) but not in the DT condition (F(1, 16) = 0.767, p = 0.394). 15 To assess the influence of the HA provision in the SSD group, a 2-way repeated measures ANOVA (Type III) with within-subject factors group (SSD, SSDaided) and task (ST, DT) was conducted. While the interaction (F(1, 7) = 0.074, p = 0.794) and the main effect group (F(1, 7) = 1.529, p = 0.256) remained non-significant, the main effect task (F(1, 7) = 58.443, p < 0.001) appeared to be significant. Figure 7: Amount of correctly repeated target words (in %) of the CCOLSA ST and DT condition for the SSD, BIM and SSDaided groups. The boxes extend to the interquartile range, with the median represented by a horizontal line. The whiskers extend to a maximum of 1.5 times the length of the box. Statistical significance is indicated by asterisks (*p<.05, **p<.01, ***p<.001). 3.4 CCOLSAcosts The distributions of the CCO are shown in Fehler! Verweisquelle konnte nicht gefunden werden.8. The median CCO of the SSD group was 4.6 dB and 5.9 dB in the BIM group. For the SSD data, a deviation from the normal distribution was revealed (SW-Test, p = 0.045), while normal distribution could be assumed for the BIM group (SW-Test, p = 0.2). The paired comparison revealed a statistically significant difference in CCO between the SSD and BIM groups (U = 11, Z = 2.579, p = 0.01). 16 Figure 8: CCO of the SSD (green) and BIM (blue) group. The boxes extend to the interquartile range, with the median represented by a horizontal line. The whiskers extend to a maximum of 1.5 times the length of the box and outliers are marked with ’x’. Statistical significance is indicated by asterisks (*p<.05). 17 4 Discussion This study investigated the speech recognition abilities of SSD and BIM patients in different situations. Therefore, SRTs were measured using different maskers and different spatial scenarios from which SRM values were extracted. Additionally, the speech recognition for complex situations was assessed using a multi-talker test which has never been applied to CI users before and which also captures the influence of different attention situations. Furthermore, the hearingrelated cognitive performance in terms of CCO was assessed. 4.1 Speech Recognition For the OLSA reference condition, S0N0 with olnoise, the SRTs were −8.5 dB SNR for the SSD group. This corresponds to SRTs of normal hearing listeners (Wagener et al., 1999b) (−7.1 dB SNR) and was also expected according to Williges et al. (2019) (approx. −6.0 dB SNR) for the SSD group. SRTs of −4.5 dB SNR were measured for the BIM group, which is 4 dB higher (poorer speech recognition) than the SSD group’s SRTs and is also in line with the findings of Williges et al. (2019) (approx. −4.0 dB SNR). This indicates that for SR without spatially separated sound sources, better ear (BE) performance is decisive. The BE can be determined through comparison of the S0N90 conditions within a group. For the SSD group, the acoustical hearing ear, and for the BIM group the electrical hearing ear appeared to be the BE. In scenarios with spatially separated sound sources SRTs of −10.4 dB SNR (S0N90ac) and −12.5 dB SNR (S0N90CI) were measured for the SSD group in the olnoise condition. This corresponds to the expectation according to Williges et al. (2019), which stated that in the SSD group, better SR was to be expected with the masker on the CI side. The measured results are comparable to those determined by Williges et al. (2019), who measured SRTs of approx. −10.0 dB SNR in the S0N90CI condition. However, the SSD group in this study showed 6.0 dB lower SRTs (better speech recognition) for the S0N90ac condition than the SSD group of Williges et al. (2019) (approx. −4.5 dB SNR). When comparing SRTs for conditions with spatially separated sound sources, substantial differences can arise due to variations in loudspeaker setups and room acoustics, as illustrated by the reference curves reported by Winkler et al. (2025). Normal hearing listeners can reach SRTs of approx. −17.0 dB SNR in these scenarios and therefore distinctly better SR (Beutelmann & Brand, 2006). With SRTs of −10.7 dB SNR (S0N90ac), the BIM group reached comparable results to the SSD group, but no improvement was observed when presenting the noise on the CI side. Presenting the noise from the CI side leads to a degradation of speech recognition and SRTs of −7.0 dB SNR (S0N90CI). This is also in line with the findings of Williges et al. (2019), which stated that the electrical hearing ear is BE for the BIM group. Using the FIT as masker in a S0N0 scenario, SRTs of −16.1 dB SNR were observed in the SSD group, which was 10.2 dB lower (better speech recognition) than the BIM groups SRTs of −5.9 dB SNR. Grossmann et al. (2016) measured SRTs of −1.6 dB SNR for SSD patients using 18 the OLSA and a varying male two-talker babble noise in a S0N0 scenario. Hence, the SSD group in this study exceeded the expectations demonstrating lower SRTs (better speech recognition) than the SSD group in Grossmann et al. (2016), which could be explained due to the difference between the used maskers. The BIM groups SRTs for the same condition exceeded the expectations according to Cullington and Zeng (2011), who measured SRTs of approx. 1 dB SNR, using sentences of the HINT database with one female interfering talker from the Institute of Electrical and Electronics Engineers (IEEE) sentence material. For the S0N90CI condition SRTs of −18.2 dB SNR for the SSD group and −6.0 dB SNR for the BIM group were observed. The SSD group again demonstrated clearly lower SRTs (better speech recognition) than the BIM group, with a difference of 12.2 dB and again exceeded the expectation according to Grossmann et al. (2016), who observed SRTs of −8.8 dB SNR for the S0N90CI condition. The BIM participants of Gifford et al. (2014) achieved SNR50 scores of 2.0 dB SNR for the same condition, revealing a difference of approx. 4 dB to the S0N0 condition, which is different from the BIM group in this study who reached comparable SRTs to the S0N0 condition. A similar trend was observed for the S0N90ac condition, where SRTs of −17.1 dB SNR were observed for the SSD group, which is 8.1 dB lower (better speech recognition) than the −9.0 dB SNR of the BIM group. Grossmann et al. (2016) observed SRTs of −3.1 dB SNR (worse speech recognition) for their SSD group, thus the SSD group in this study again exceeded the expectations. In contrast to this the BIM group of Gifford et al. (2014) demonstrated similar results like the BIM group of this study, showing SNR50 scores of approx. 8.0 dB SNR. In the diffuse cafeteria condition (S0Ndiff), the SSD group demonstrated lower SRTs of 0.2 dB SNR (better speech recognition) compared to the BIM group with 3.6 dB SNR. Not only is there a significant difference between the two groups, but both groups also exhibit higher SRTs (poorer speech recognition) than in the stationary noise conditions. To analyze the benefit of spatial separation of sound sources, the SRM was extracted from the individual SRTs of each participant. The highest SRM observed in this study was 3.9 dB with olnoise for the SSD group and 6.0 dB for the BIM group. The SRM for the SSD group aligns with the results of Williges et al. (2019), where the SSD group showed an SRM of 3.4 dB. However, in their study, the BIM group demonstrated a lower SRM of only 2.3 dB, which contrasts with the higher SRM observed for the BIM group in this study. The SRM reported by Williges et al. (2019) was significantly different between SSD and BIM listeners. Since no significant differences in SRM were found between groups in this study, these results will not be discussed further. 4.2 Masker influence Statistical analysis revealed a significant difference between all maskers for the SSD group, but not for the BIM group. There was no significant difference between the maskers olnoise and FIT in the BIM group. The olnoise primarily causes energetic masking (EM), which interferes with speech recognition by overlapping the target speech signal. In contrast, the FIT induces 19 informational masking (IM), where the competing talker’s speech content distracts the listener and makes it harder to focus on the target speech. In order to separate two competing talkers, it is crucial to detect and differentiate between the talkers’ fundamental frequencies, like differentiating between the male OLSA target talker and the female interfering talker (FIT masker). Individuals with hearing loss often have a poorer frequency selectivity compared to normal hearing listeners, while CIs struggle with the presentation of harmonic signal components (Oxenham, 2008). Both factors - impaired frequency selectivity and problematic harmonics presentation - are found in the BIM group and could explain why none of both maskers used in this study showed a significant influence on the SR of the BIM group. Pyschny et al. (2011) investigated the influence of IM on the SR of BIM patients but used a different procedure. They used OLSA sentences as target speech, with a fixed level and SNR, to determine the correct SR in percentage. As maskers they used OLSA sentences, but with increased fundamental frequency. The measurements were conducted in a S0N0 scenario for the conditions HA only, CI only and bimodal (HA + CI). No improvement for the separation by different fundamental frequencies of target and interfering talker was found through the bimodal listening mode compared to CI or HA only. Müller and Lang-Roth (2021) used a similar procedure and stimuli like Pyschny et al. (2011) to investigate the influence of IM on SSD patients and additionally tested the influence of different spatial scenarios. For the S0N0 condition, no significant improvement was observed comparing the bilateral listening mode to the normal hearing ear alone, indicating that separation of different talkers primarily depends on the performance of the normal hearing ear. Vongphoe and Zeng (2005) compared the talker recognition based on temporal signal information between acoustical and electrical hearing and found worse performance for CI users. This also supports the assumption that, since the CI ear is the BE for the BIM group, this type of masker has only a minor influence. The results are in line with the findings of this study, demonstrating a significant difference between olnoise and FIT only for the SSD group and not for the BIM group. 4.3 Attention eects and cognitive performance An important part of this study was to investigate whether CCOLSA could be measured with CI patients. It was shown that the measurements were feasible for both SSD and BIM patients. The participants’ age ranged from 24 to 81 years; thus, it can be assumed that it can also be conducted with other CI patients in the same age range as long as the measurement of other speech in noise tests is feasible for them. The SR in complex listening situations was significantly different between both groups. Furthermore, only the SSD group had an advantage in the SA task, which could not be observed in the BIM group. A HA provision on the contralateral ear had no influence on the SR performance or the SA benefit of the SSD group. Significantly higher hearing-related cognitive performance in terms of CCO was measured for the SSD group compared to the BIM group. 20 All groups reached a word scoring close to 50% correct (SSD: 54.1%, BIM: 52.5%, SSDaided: 42.5%) for the DT condition, which was expected due to the usage of the individual overlap times determined in the adaptive condition. Hence, each participant group demonstrated plausible results for the CCOLSA. The word scoring results of the ST condition (SSD: 84.1%, BIM: 59.1%, SSDaided: 80.0%) are in line with the reference condition (S0N0) of the OLSA with olnoise (SSD: -8.5 dB, BIM: -4.5 dB), showing better SR for the SSD than for the BIM group. Through comparison of the ST and DT results, a selective attention benefit was observed for the SSD and SSDaided group, which could not be observed for the BIM group. This difference could be explained by the difference of the acoustical hearing side. According to Shinn-Cunningham and Best (2008), there is a decisive difference between the benefit through SA in normal hearing listeners and the functionality of HA features. Taking into account that the HA provision of the acoustical hearing ear in the SSDaided group showed no significant influence on the performance of the SSD participants, the difference between the groups is probably not based on the additional signal processing of the HAs. Rather, it can be assumed that it is determined by the hearing loss. The CCO values were calculated based on the individual overlap time determined in the adaptive CCOLSA and provide information about the hearing related cognitive performance of the participants. With CCO values of 4.6 dB the SSD group reached significantly lower (better hearing related cognitive performance) CCO than the BIM group with 5.9 dB. This study was the first time CCO were measured in any CI patient group, so there is a lack of reference data using the same measure. When introducing the adaptive CCOLSA, Heeren et al. (2023) provided CCO scores for normal hearing listeners, hearing impaired listeners without HA provision (unaided non-HA users) and unaided HA users. Posthoc, only the difference between normally hearing listeners and HA users was found to be significant by Heeren et al. (2023). The normally hearing listeners achieved CCO of approx. 5.0 dB, after which the SSD group in this study appeared to possess a slightly better performance than the normal hearing listeners tested by Heeren et al. (2023). With CCO of approx. 6.1 dB for the unaided HA users and approx. 5.8 dB for the unaided non-HA users group, both demonstrated comparable CCO to the BIM group in this study. Since the performance on the CI side was matched between both groups, the difference in hearing related cognitive performance appears to be determined by the acoustical hearing side. Hearingimpaired listeners can perform similarly to normal hearing listeners in tasks including SR, but they have to dedicate more cognitive resources on the task (McCoy et al., 2005). The results of this study show that even with the combination of an electrically hearing and an acoustically normally hearing side, the performance of SSD patients described by the CCO is similar to that of normal hearing listeners. This contrasts with the finding that the performance of BIM subjects is similar to that of unaided non-HA users and unaided HA users. 21 Since SSD patients gain an advantage through SA and can reach CCO comparable to normally hearing listeners results, SRM appears to be the only difference between SSD patients and normal hearing listeners. SRM is measured most commonly in scenarios using stationary noise sources. Hence, the question arises whether normal hearing listeners show an advantage in complex listening situations compared to SSD patients. Puglisi et al. (2021) investigated the influence of reverberation time and different noises on the SR of normal hearing listeners in real, complex acoustical scenarios. One of the key findings was that even for these situations, SRM of up to 3 dB could be observed. It is noticeable that significant SRM for EM only appeared for low reverberation times and for IM only for large reverberation times. Therefore, it can be concluded that SRM also plays a role in complex listening situations. It can be assumed that normal hearing listeners have an advantage over SSD subjects in these situations. 4.4 Study limitations One issue when looking at BIM patients is the high variability of measurement results inside the group, due to the high amount of possible hearing losses. It is difficult to precisely estimate the influence of hearing loss. However, the results of this study state that the CI ear is the BE for BIM patients, as also found by Williges et al. (2019) and thus the contribution of the acoustical ear to SR is rather small, especially for complex listening situations. Based on the recommendations of the G-BA (2021), an increase of 20% of MWR is enough for a successful HA provision. Hence, even with perfectly fitted HAs, big differences between the subjects are possible. No restrictions were made regarding the SR of the acoustical hearing ear. For better interpretation and analysis of future results, it could be useful to define subgroups inside the BIM group, based on the hearing loss or the SR with HA only. When interpreting the influence of HAs on SR by comparing SSD and SSDaided results, it should be noted that no acclimatization period was provided and that measured pure tone thresholds do not necessarily indicate the need for HA provision in most SSD patients. Another aspect, that should be taken into account when interpreting the results, is the small number of participants, leading to a rather poor statistical power. However, for a study including CI patients and carried out at a single institution, the number of participants is rather large. As a comparison Williges et al. (2019) recruited eight SSD and eight BIM patients from two different institutions. While our SSD group’s results are comparable to the findings of other studies, the BIM group showed predominantly better results compared to literature. This may be attributed to the selection of the participants. Based on both the measurement results and the subjective assessment of the responsible audiologist, all participants in this study were rated as high performers. Therefore, the generalization of the findings to the entirety of SSD and BIM patients should be approached with caution. 22 5 Conclusions This study systematically assessed speech recognition performance for situations with different maskers as well as for dynamic complex listening situations, for SSD and BIM patients. Furthermore, the hearing-related cognitive performance of the participants was assessed. The main conclusions are: 1. CCOLSA is feasible for both SSD and BIM patients. 2. Comparable monosyllabic word recognition with the CI does not indicate comparable speech recognition for complex listening situations. 3. SSD patients reached lower SRTs (better speech recognition) than BIM patients for the maskers olnoise, FIT and diffuse cafeteria noise. 4. SSD patients gain an advantage through selective attention in complex listening situations; BIM patients do not. 5. SSD patients showed better hearing-related cognitive performance than BIM patients. Speech recognition in everyday life is influenced by a variety of factors. The successful measurement of the CCOLSA revealed new differences between SSD and BIM patients. As the results of this study show, current standard procedures, such as the measurement of the FBE and the use of stationary signals, do not sufficiently capture these effects. Therefore, an extension to include tests that incorporate speech-in-speech conditions and assess cognitive abilities is recommended. However, further research is needed to shed more light on the differences between SSD and BIM patients. In particular, the influence of hearing loss in BIM patients seems to be a promising approach and deserves further investigation. 23 References Ahrlich, M. (2013). Optimierung Und Evaluation Des Oldenburger Satztests Mit Weiblicher Sprecherin Und Untersuchung Des Effekts Des Sprechers Auf Die Sprachverständlichkeit. Optimization and evaluation of the female OLSA and investigation of the speaker`s effects on speech intelligibility. Bachelor Thesis. Alfakhri, M., Campbell, N., Lineton, B., Rowan, D., & Boyle, P. (2024). International survey of bimodal hearing and bilateral cochlear implant service provision for adults. Cochlear Implants International, 25(4), 260–274. https://doi.org/10.1080/14670100.2024.2413267 Arndt, S., Laszig, R., Aschendorff, A., Beck, R., Schild, C., Hassepass, F., Ihorst, G., Kroeger, S., Kirchem, P., & Wesarg, T. (2011). Einseitige Taubheit Und Cochlear-implant-Versorgung: Audiologische Diagnostik Und Ergebnisse. HNO, 59(5), 437–446. https://doi.org/10.1007/s00106-011-2318-8 Arndt, S., Laszig, R., Aschendorff, A., Hassepass, F., Beck, R., & Wesarg, T. (2017). CochleaImplantat-Versorgung von Patienten Mit Einseitiger Taubheit Oder Asymmetrischem Hörverlust. HNO, 65(Suppl 2), 98–108. https://doi.org/10.1007/s00106-016-0297-5 Ausili, S. A., Agterberg, M. J. H., Engel, A., Voelter, C., Thomas, J. P., Brill, S., Snik, A. F. M., Dazert, S., Van Opstal, A. J., & Mylanus, E. A. M. (2020). Spatial Hearing by Bilateral Cochlear Implant Users With Temporal Fine-Structure Processing. Frontiers in Neurology, 11, 915. https://doi.org/10.3389/fneur.2020.00915 Avan, P., Giraudet, F., & Büki, B. (2015). Importance of Binaural Hearing. Audiology & neurootology, 20 Suppl 1, 3–6. https://doi.org/10.1159/000380741 Beutelmann, R., & Brand, T. (2006). Prediction of Speech Intelligibility in Spatial Noise and Reverberation for Normal-Hearing and Hearing-Impaired Listeners. The Journal of the Acoustical Society of America, 120(1), 331–342. https://doi.org/10.1121/1.2202888 Brand, T., & Kollmeier, B. (2002). Efficient Adaptive Procedures for Threshold and Concurrent Slope Estimates for Psychophysics and Speech Intelligibility Tests. The Journal of the Acoustical Society of America, 111(6), 2801–2810. https://doi.org/10.1121/1.1479152 30 Wagener, K. C., Hansen, M., & Ludvigsen, C. (2008). Recording and Classification of the Acoustic Environment of Hearing Aid Users. Journal of the American Academy of Audiology, 19(4), 348–370. https://doi.org/10.3766/jaaa.19.4.7 Williges, B., Wesarg, T., Jung, L., Geven, L. I., Radeloff, A., & Jürgens, T. (2019). Spatial Speechin-Noise Performance in Bimodal and Single-Sided Deaf Cochlear Implant Users. Trends in hearing, 23, 2331216519858311. https://doi.org/10.1177/2331216519858311 Winkler, A., & Holube, I. (2014). Was Wissen Wir Über Den Freiburger Sprachtest? Zeitschrift für Audiologie (Audiological Acoustics), 53, 146–154. Winkler, A., Warkentin, L., Denk, F., Husstedt, H., Sankowksy-Rothe, T., Blau, M., & Holube, I. (2025). Reference Speech-recognition curves for a German monosyllabic test in noise: Effects of loudspeaker configuration and room acoustics. International Journal of Audiology, 64(7), 695–704. https://doi.org/10.1080/14992027.2024.2401519 Zeitler, D. M., Dorman, M. F., Natale, S. J., Loiselle, L., Yost, W. A., & Gifford, R. H. (2015). Sound Source Localization and Speech Understanding in Complex Listening Environments by Single-sided Deaf Listeners After Cochlear Implantation. Otology & neurotology: official publication of the American Otological Society, American Neurotology Society [and] European Academy of Otology and Neurotology, 36(9), 1467– 1471. https://doi.org/10.1097/MAO.0000000000000841