Tech Topic | March 2019 Hearing Review
A multinational survey of hearing aid users shows substantial advantages and high satisfaction rates for the new Widex EVOKE hearing aid. Additionally, the device’s SoundSense Learn feature was shown to provide real-time personal sound customization that would not have been achievable via a conventional fitting rationale.
Hearing happens in real life, while the hearing aid user is getting their car washed, picking up their kids, or watching their favorite TV show. That is why the new Widex EVOKE hearing aid is designed to meet the user’s listening intention in each specific real-life moment, through a combination of automation and personalization. Automation meets users’ needs in most situations, but cannot meet listening expectations for all users all the time, making personalization in the moment key. In EVOKE, this personalization is delivered by SoundSense Learn, a real-time machine-learning algorithm that quickly and intuitively adjusts hearing aid parameters to meet the listening intent of the user.
To understand the benefits of this unique combination of automation and personalization, we conducted a large-scale multinational survey to investigate users’ experiences of Widex EVOKE compared to their own hearing aids in a range of situations. The survey shows substantial advantages for EVOKE, with particularly large improvements for noisy situations, which are notoriously difficult for hearing aid users.1-3 The survey also shows advantages of the SoundSense Learn feature, which uses real-time machine learning to allow the user to personalize sound in the moment.
The advantage of EVOKE in noisy situations is particularly important because speech intelligibility in noise is a huge challenge for people with hearing loss. Even with hearing aids, hearing loss may force people to avoid situations with too much noise. And these situations are not rare or special; they are everyday situations of real-life hearing: talking to a partner while driving, going to a restaurant with friends, eating lunch in the cafeteria at work, celebrating family occasions. Withdrawing from such social interactions is harmful to both health and quality of life, as detailed by the National Council on Aging’s report on consequences of untreated hearing loss.4
The importance of the noise problem also appears repeatedly in MarkeTrak surveys, where noisy situations consistently get the lowest satisfaction ratings from hearing aid users.1 At the same time, performance in noisy situations is rated as the most important hearing aid parameter in a recent Hearing Tracker survey.5 This means that intelligibility in noise is simultaneously the most dissatisfying and the most important hearing aid performance parameter for users—a strong motivator in the development of EVOKE.
Automation in EVOKE is driven by Fluid Sound Technology. First, the Fluid Sound Analyzer automatically categorizes the acoustic environment. This is done with increased accuracy, driven by a doubling of analyzer training during the development phase. Moreover, the analyzer has more choices when classifying sound with two new sound classes. This allows the system to incorporate an even larger number of commonly encountered listening environments with 11 sound classes: Quiet, Urban, Transport, and Party with and without speech, an updated categorization of music into Classical or Contemporary, and the entirely new Social sound class, which bridges the gap between the Party and Quiet sound classes with settings optimized for small gatherings. In turn, the Fluid Sound Controller has access to more hearing aid parameters than previous models when it makes automatic adjustments. These parameters include noise reduction, directionality, and compression.
To guide personalization, the SoundSense Learn function employs user data and machine learning to allow the hearing aid to learn about the individual user’s preferences in real time. This feature was developed because we know that even with the best-fitted hearing aids, users may sometimes find themselves in situations where the sound is not suited to the specific situation and their specific listening intention. SoundSense Learn allows tailoring of the sound in such situations, without requiring the user to understand complex controls.
SoundSense Learn varies gain settings in three frequency bands based on user input. Users simply indicate their relative preference for two gain settings, A or B, using their EVOKE smartphone app. The process starts with a comparison of one pair of A and B settings, after which a real-time machine-learning algorithm continually determines the next A-B comparison, converging towards a final setting that best matches the user’s preference. The user should achieve best convergence after maximally 20 trials, but user reports suggest that preferable sound is often reached after just a few comparisons. The result is a personalized gain setting that can be used just for that moment or saved as a program for later use. We have seen in both internal6 and external double-blind lab studies (manuscript in preparation) that allowing users to personalize sound with SoundSense Learn significantly increases perceived sound quality and, in some cases, listening comfort. For more detailed descriptions of how SoundSense Learn works, readers are referred to Nielsen, Nielsen & Larsen7 and Townend, Nielsen & Ramsgaard.8
In addition to allowing the user to personalize sound in the moment to match their listening intent, the intelligent use of data and machine learning is also central for Widex research and development, where we use anonymized user data to further develop the SoundSense Learn feature. We also consider these data in this article to learn more about real-life hearing aid personalization.
The user survey was conducted across 9 different countries (Canada, China, the Czech Republic, Hungary, Italy, Japan, Portugal, the UK, and the US). A total of 118 hearing aid users were fitted with Widex EVOKE 440 Fusion2 hearing aids. They compared these to their own hearing aids (of various brands) in questionnaires on specific aspects and specific situations of hearing aid use. The survey ran over 7 weeks in a cross-over design, giving each user time to experience the Widex EVOKE hearing aids and compare them to their own hearing aids.
Participants. The 118 survey participants were adults recruited by their hearing care professionals (HCPs). They had all types and configurations of hearing loss within the fitting range of the Widex M-receiver (Figure 1). Participants were required to have internet access in order to complete the questionnaires online, and to be smartphone users in order to use the EVOKE app. The participants were between 18 and 89 years of age (mean age = 63, SD = 15 years), with 64 working participants and 54 non-working (mainly retired) participants. There were 39 females and 79 males. All participants signed a consent form before participating in the survey.
Survey design and procedure. The survey used a cross-over design in which hearing aid users answered parallel questionnaires at four different times—two about EVOKE and two about their own hearing aids. The survey process started at Week 0 with a visit to the clinic, where the participant answered a questionnaire about their satisfaction with their own hearing aids and were fitted with EVOKE hearing aids using the Widex EVOKE fitting rationale. Participants then wore the EVOKE hearing aids for 3 weeks, before answering a questionnaire about their experience with the new device at Week 3. This was followed by 2 weeks of wearing their own hearing aids, and then answering another questionnaire about own hearing aids at Week 5. For the final 2 weeks, participants again wore EVOKE and finished with a questionnaire about them at Week 7. These repeated answers to the questionnaires provided more reliable data about hearing aid satisfaction than single measurements.
The questionnaires were modeled on the MarkeTrak surveys1,9 and focused on user satisfaction with different aspects of hearing aid use (eg, physical fit, ability to hear soft sounds, and comfort while listening to loud sounds) and in different situations (eg, conversations in noise, traffic, or use in restaurants). Users rated satisfaction on 7-point Likert scales going from very dissatisfied (assigned a value of 1) to very satisfied (value of 7). All questionnaires contained parallel questions, except that the EVOKE questionnaires in Weeks 3 and 7 included questions about the EVOKE app and the SoundSense Learn functionality, which were not relevant for participants’ own hearing aids.
The data were collected in online questionnaires through www.smart-trial.co.
Overall satisfaction ratings. Across questions, satisfaction with the EVOKE hearing aids was consistently higher than with own hearing aids by at least one point on the 7-point scale, and in many cases—specifically situations that included noise—substantially more. We also saw ratings of own hearing aids decline in many cases from Week 0 to Week 5 after users have experienced the performance of the EVOKE hearing aids. This suggests that even though people at the outset reported reasonable satisfaction with their existing hearing aids (although less so in noisy situations), they also experienced a difference following their use of EVOKE, leading to a decline in satisfaction with own hearing aids.An example of this pattern can be seen when we consider users’ answers to the question, “Overall, how satisfied are you with the current hearing aids you have worn?” (Figure 2). The left panel shows the advantage for EVOKE in mean ratings, and the right panel shows that the majority of individual EVOKE ratings are either “Satisfied” or “Very satisfied,” while the majority of own hearing aid ratings are “Somewhat dissatisfied” or “Somewhat satisfied.”
We tested these results statistically in the regression model summarized in Table 1. This compares all questionnaire times to Week 5, and we see highly significant differences (p < 0.0001) to all three other questionnaire times, with the largest differences between the EVOKE ratings in Weeks 3 and 7 on the one hand and the Week 5 own HA ratings on the other. In addition, there were two other interesting effects: First, those participants whose own hearing aids were various Widex models were more satisfied overall, both with their existing hearing aids and with the new devices. Second, participants who indicated that they wore the hearing aids longer were generally more satisfied.
Turning to another general satisfaction parameter, sound quality, we also see enhanced satisfaction for EVOKE, as illustrated in Figure 3. This indicates that the sound quality is superior, with both higher mean ratings and many more participants who are “Satisfied” or “Very satisfied” with the sound quality of the new hearing aids.
Satisfaction in noisy situations. The survey contains a range of questions about situations that involve some degree of noise: use in noisy backgrounds, while following conversations in noise, in large groups, in restaurants, for conversations during transport, and in a noisy street. These questions all address everyday, real-life hearing situations where noise plays a large role—situations of great importance to most hearing aid users. For these noise-related situations, the median EVOKE ratings are consistently at least 2 points above the median ratings of own hearing aids, with the EVOKE median either “Satisfied” or “Somewhat satisfied,” and the own hearing aid median either “Neutral” or “Somewhat dissatisfied.”
This pattern is clear in Figure 4: There are markedly different distributions for own hearing aids (grey bars) compared to EVOKE (dark purple bars). By far, the most frequent answer for EVOKE in noisy situations is “Satisfied,” while for own hearing aids, the bulk of the ratings are “Very dissatisfied,” “Dissatisfied,” or “Somewhat dissatisfied.”
To assess overall satisfaction in noise, we calculated the mean rating across the six noise questions. An overwhelming majority (88%) of participants are more satisfied with the EVOKE performance across noisy situations compared to their own hearing aids. If we only consider the subset of users whose own hearing aids were not Widex models, that number becomes even higher, with 94% of users being more satisfied with Widex EVOKE than with their own non-Widex hearing aids. As mentioned above, satisfaction in noise with EVOKE meets the highest-ranked need for hearing aid users.5
This advantage in noisy situations is likely due to a combination of EVOKE features. The linear input dynamic range in EVOKE enables an uncompromised sound in some of the most challenging listening situations. The upper input limit of 113 dB SPL in the analog to digital converters ensures that incoming sound is not compressed or distorted before reaching the digital signal processing (DSP) of EVOKE. A clean and uncompromised signal enables the DSP features to perform as they are designed.
For example, a noise reduction feature must be able to separate speech from noise in the signal in order to be effective. This becomes much more challenging, if not impossible, if the signal is distorted due to input signal limits. We have seen evidence that this high upper limit allows hearing-impaired participants to understand almost as much speech as normal-hearing subjects in the same noisy, challenging listening situation.10
An additional advantage is provided by the Speech Enhancer algorithm, which identifies the need for additional gain and reduction of noise by applying the Speech Intelligibility Index (SII). Adjustments are based on an instantaneous analysis of the environment, the patient’s hearing loss, and the goal of emphasizing speech over noise, while incorporating real-time binaural inter-aural synchronization. This has been shown to lead to a 2 dB improvement of speech in noise ability.11
Satisfaction with soft sounds. Another central aspect to look at is the ability to hear soft sounds. This is a signature of Widex hearing aids for several reasons, including the low compression threshold (CT) that provides extra gain to low-input sounds and ensures consistent audibility for soft speech.12 A low CT is beneficial in a variety of situations, including listening in a quiet place, and understanding speakers and languages that “fall off” in amplitude towards the end of a sentence.13 Additional features that contribute to Widex audibility for soft sounds include the input dynamic range that begins as low as 5 dB SPL; Variable Speed Compression to maintain audibility for soft sounds following loud sounds; and Soft Level Noise Reduction, which removes constant low-level noises such as room ventilation and PC fans. These features contribute to enhance audibility for soft speech and to the overall higher satisfaction with the ability to hear soft sounds with EVOKE. Figure 5 illustrates higher mean ratings for EVOKE, and many more ratings of “Satisfied” and “Very satisfied”; the most frequent answer for own hearing aids is “Somewhat dissatisfied.”
Users Recommend SoundSense Learn
A key feature of the EVOKE hearing aids is SoundSense Learn, which uses real-time machine learning to allow the user to personalize sound. This provides an intuitive tool for those situations where the user experiences difficulties in the moment, or where listening intent differs from what is anticipated in a specific environment. With SoundSense Learn, users can address this challenge without requiring a fine-tuning appointment and, more importantly, without missing out on real-life moments. It is a foundational aim of EVOKE to meet individual listening intent, a variable that has not previously been fully addressed in amplification systems.
In the survey, we asked those participants who indicated that they had used SoundSense Learn if they felt it had helped them improve a listening situation, and if they would recommend it to others. Answers here are overwhelmingly positive, with 38 out of 53 participants (72%) in Week 7 reporting an improvement after using SoundSense Learn, and 42 out of 53 (80%) reporting a willingness to recommend the feature to others (Figure 6).
It is of note that not all survey participants used SoundSense Learn, but this is exactly what we would expect: In most cases, the automation in EVOKE provided performance, audibility, and sound quality that made adjustments unnecessary. Additionally, for EVOKE, we have refined the fitting rationale used for the first fit,14 which also reduces the need for further adjustment. However, for those situations where personalization is relevant, SoundSense Learn provides users with a powerful and intuitive tool.
While the survey data provided information about user satisfaction with SoundSense Learn, we were also crucially interested in how a broader group of wearers use SoundSense Learn in real life. To this end, we explored the anonymous data that are collected when users create and use their individual SoundSense Learn programs, separately from the survey. From this, we estimated how many were regular users of SoundSense Learn, how much they used their personalized SoundSense Learn settings, and what those settings looked like in terms of the gain settings in the three frequency bands that can be manipulated.
Looking first at the number of users, approximately 1 in 5 who had access to SoundSense Learn used it. This number indicates that the automation in EVOKE meets the needs for most users in most situations, but that a substantial group of users also benefit from being able to personalize sound. Interestingly, the proportion of SoundSense Learn users has increased substantially since this was investigated in 2018,15 indicating an increase in popularity.
To learn more about this group of SoundSense Learn users, we considered a sample of around 13,000 unique users, who have created over 33,000 individual programs, for an average of approximately 2.5 per person. Around half of the users have saved one program, while the other half have saved multiple programs for use across different situations. The vast majority of the programs get used more than once, with more than half the programs used at least 5 times. In addition to these saved programs, EVOKE wearers may also use SoundSense Learn to adjust sound in the moment without saving the settings. This use is likely to be frequent, but is not included in our sample.
The gain adjustments that users make with SoundSense Learn are illustrated in Figure 7, where each dot represents a program, showing the gain adjustment for the three different frequency bands on the three axes. Figure 7 shows only a representative subset of the data in order to be able to see a distribution of the programs (but the same pattern is observed if we consider the full sample). The programs are distributed all over the three-dimensional space, with no apparent clusters, indicating that the adjustments made are not ones that could be addressed by changes to the fitting rationale. This pattern is a strong indication that the programs the users created with SoundSense Learn are truly personal, with settings that are impossible to predict.
From a user perspective, each dot illustrates a situation where the user was able to accommodate a unique listening intention and preference in real time. From an HCP perspective, each dot represents a potential need for fine-tuning that did not require a visit to the clinic because the user had access to SoundSense Learn. Initial fitting and more general fine-tuning remain specialized tasks for the HCP, but SoundSense Learn allows the end user to address problems in highly specific situations—problems which the HCP may be challenged to solve based on what limited information the end user can provide in retrospect. This is likely to increase end-user satisfaction and free up HCP resources for other tasks, including the kind of counseling that may in turn increase end-user satisfaction even more.
Every moment of real-life hearing is different, and every user has a specific intention for what they want to focus on hearing in the moment. This calls for a hearing aid that can adjust to every situation and to every listening intention. Widex EVOKE does this automatically with its refined Fluid Sound Technology, and the results are reflected in this user survey, where EVOKE gets higher ratings both on general satisfaction and across a range of different situations.
An especially strong point of the new hearing aids is performance in noisy situations, where 88% of users are more satisfied with EVOKE than with their own hearing aids. Noisy situations are both the most challenging and the most important areas for users, so with its improved performance in noise, EVOKE addresses exactly those situations that are most critical in real life. This very substantial increase in satisfaction in noisy situations—which goes up to 94% for users whose existing hearing aid was not a Widex model—is a strong argument for using EVOKE.
Another strong argument for the new hearing aid is the ability to personalize sound in the moment. SoundSense Learn is popular both in the survey data and the user data collected when the feature is used. The data clearly show that the adjustments made with SoundSense Learn are so varied that they could not be covered by changing the fitting rationale; therefore, it provides a tool that allows each individual user to make adjustments that suits his/her life and listening intention. All in all, the fine balance between automation and personalization makes EVOKE a hearing aid that truly works across real-life situations and helps users fully participate in their everyday lives, exactly as it was intended to.
The authors are deeply grateful to the following clinics for participating in the survey: From the US: Associated Audiologists; Audio Acoustics; Center for Audiology; Clark Hearing; ENT Hearing Associates; Estes Audiology; Family Hearing; Holland Doctors of Audiology; Pacific Hearing Services; San Francisco Audiology; Texas ENT Center. From Canada: Alderney Hearing Centre; Audiology Clinic of Northern Alberta; Bentley Hearing Services; Davidson Hearing Aid Centres; Grand River Hearing & Tinnitus Centre; Harp Hearing Care; Marco Hearing Health Centre; Newlife Hearing Health Centre; North Bay Audiology Clinic; Sackville Hearing Centre. From the Czech Republic: MUDr Jan Adam, Macadam, Liberec; MUDr Libor Cerny, Vseobecna Fakultni Nemocnice, Praha; MUDr Radan Havlik, Audio-fon Centre, Brno; MUDr Stepanka Havlova, ORL, Foniatrie, Sluchová Protetika, Plzen; MUDr Ivan Jedlicka, Widex, Praha; MUDr Monika Vohlidkova, Fakultní nemocnice, Plzen. From China: Hangzhou Huier; Fenyang Vision and Audition Center (affiliated to Shanghai EENT Hospital); Shanghai Nordic Sound; Shanghai Ninth People’s Hospital (affiliated to Shanghai Jiaotong University Medical College); Zhengzhou Boshi; Chongqing Logic Sound; Chengdu Yier. From Hungary: clinics in Budapest: Pécs; Szeged; Debrecen; Szombathely; Gyomaendr?d; and Kalocsa. From Italy: Instituto Acustico Pontico Pontoni; Audiologica SM; Italfon; Otomedical; Acustica M. From Japan: Bloom Ginza; Bloom Yokohama West; Mimi HA; Matsuzakaya Ueno HA Salon; Bloom Matsudo; Bloom Nakano. From Portugal: Widex Cascais; Widex Guimarães; Widex Braga; Widex Faro; Widex Lisboa. From the UK & Ireland: Connect Hearing Stillorgan; Charnwood Hearing Centre; DigiClear Hearing & Balance; Kent Hearing; McCreesh Hearing.
Abrams HB, Kihm J. An introduction to MarkeTrak IX: A new baseline for the hearing aid market. Hearing Review. 2015;22(6):16-21.
Gygi B, Hall DA. Background sounds and hearing-aid users: A scoping review. Int J Audiol. 2016;55(1):1-10.
McCormack A, Fortnum H. Why do people fitted with hearing aids not wear them? Int J Audiol. 2013;52(5):360-368.
National Council on the Aging. The Consequences of Untreated Hearing Loss in Older Persons. https://www.ncoa.org/wp-content/uploads/NCOA-Study-1999.pdf. Published May 1999.
Bailey A. Fresh data from Hearing Tracker’s hearing aid matching system. LinkedIn. https://www.linkedin.com/feed/update/urn:li:activity:6448648115971989504/. Published November 2018.
Townend O, Nielsen JB, Balslev D. SoundSense Learn—Listening intention and machine learning. Hearing Review. 2018;25(6):28-31.
Nielsen JB, Nielsen J, Jensen BS, Larsen J. Hearing aid personalization. Paper presented at: 27th Conference on Neural Information Processing Systems (NIPS); December 5-10, 2013; Lake Tahoe, Nev. http://eprints.gla.ac.uk/119556/1/119556.pdf.
Townend O, Nielsen JB, Ramsgaard J. Real-life applications of machine learning in hearing aids. Hearing Review. 2018;25(4):34-37.
Kochkin S. MarkeTrak VIII: 25-year trends in the hearing health market. Hearing Review. 2009;16(11):12-31.
Kuk F, Lau C-C, Korhonen P, Crose B. Speech intelligibility benefits of hearing aids at various input levels. J Am Acad Audiol. 2015;26(3):275-288.
Peeters H, Kuk F, Lau C-C, Keenan D. Subjective and objective measures of noise management algorithms. J Am Acad Audiol. 2009;20(2):89-98.
Kuk F. Optimizing compression: The advantages of a low compression threshold. In: Kochkin S, Strom KE, eds. High Performance Hearing Solutions, Vol 3: Hearing in Noise. Hearing Review. 1997;6(11)[Suppl]:44-47.
Chasin M. How much gain is required for soft-level inputs in some non-English languages? Hearing Review. 2011;18(10):10-16.
Schmidt E. Widex. The Widex fitting rationale for EVOKE. http://webfiles.widex.com/WebFiles/9%20502%204866%20001%2001.pdf. Published March 2018.
Balling LW, Townend O. Widex. Real-life benefits of Widex EVOKE: An early look at end-user survey results. http://webfiles.widex.com/WebFiles/WidexPress41.pdf. Published October 2018.
CORRESPONDENCE can be addressed to HR or Dr Balling at: [email protected]
About the authors: Laura Winther Balling, PhD, is an Evidence and Research Specialist and Oliver Townend, BSc, is Head of Audiological Communication at Widex A/S in Lynge, Denmark, and Wendy Switalski, AuD, MBA, is Director of Professional Development at Widex USA in Hauppauge, NY.