Sound Testing for Perceiving Change in Sound over Time in Cochlear Implants

A cochlear implant (CI) is a device that provides people with severe hearing loss or deafness the ability to acquire functional hearing by electrically stimulating the auditory nerve. CI users have a good understanding of speech in quiet situations. One of the most pressing problems is the understanding of speech in real life conditions, where there often is competing noise. Other difficult conditions include pitch variation in tonal languages, emotion in speech (prosody), sound localization and music perception (Dincer D'Alessandro et al., 2018; Liu et al., 2017; Moore, 2008; Rubinstein, 2004).Sound can basically be split into an envelope (variation of amplitude over time) and into temporal fine structure (TFS, representing variation in frequency and phase over time). CIs mainly encode the envelope and only a very limited amount of TFS. The ability to detect temporally encoded frequency information helps to detect the fundamental frequency, to perceive subtle pitch variation and to perceive timbre differences in music. With the inclusion of TFS, the sound quality of CIs would greatly improve. The TEMPORAL project is targeting this very important problem by applying AI techniques to create or modify speech encoding strategies to optimize listening in difficult situations.

The aim of this current proposal is to compare test methods for determining TFS performance, to establish detection thresholds for both CI users and normal hearing subjects and to produce normative data for the computer model and machine learning system. A wide range of psychophysical tests exists to directly and indirectly test TFS performance. Methods to test TFS directly were developed by Moore and colleagues, the TFS1 (Moore & Sek, 2009) and TFS-LF (Hopkins & Moore, 2010). Other tests focus on areas of improvement such as speech recognition in noise (Houben et al., 2014; Smits et al., 2013), pitch perception (Snel-Bongers et al., 2011; Vaerenberg et al., 2011) and music perception (Moon & Hong, 2014) and are indirect ways of detecting TFS. While these psychophysical tests have been used separately, no direct comparison has been made. Such a comparison provides insight in the relative performance of several aspects of hearing. It will also show which test is optimal for determining TFS performance. All subjects will undergo tests that are selected for their capability to inform TFS performance and provide input for computational models. In addition, tests are included that measure speech performance. Selected tests are all acoustical, meaning they are played through speakers or headphones. TFS tests compared with speech tests will provide valuable information on which tests are suitable for this purpose. Results from TFS tests provide baseline values for input in the computer models and further research. To limit contamination of the data by cognitive factors and ease of implementation into the machine learning system, we will focus on task-specific (discrimination tasks or n-alternative forced choice) psychophysical experiments.

This study is a prospective single-centred cross-sectional study. The total duration of the study is estimated to be 2.5 years. CI users and normal-hearing subjects are included as study group. Each subject will participate in a maximum of 3 non-consecutive test days. Each test day consists of a test session that lasts for approximately 3 hours, with breaks between each test and additional breaks if needed. The setting is the LUMC ENT department where there is long-standing experience with psychophysical testing.

Subjects will be asked to participate in several speech recognition or psychophysical tests. In each test, they will repeatedly hear sequences of sounds (two or three sounds) and must choose which is different. Subjects can select their choice themselves on a touch screen. Playing and choosing sounds is repeated until a certain number of turn points is reached. Subjects can then proceed to the next test, take a break or stop for the day. Sounds are played through a speaker. Loudness is set to 65 dB(A), which is the loudness of normal conversation. Each test is repeated three times. The order in which the subjects do the tests is randomized, to avoid effects of learning and fatigue. Normal hearing subjects will also take a short audiogram to check if their hearing is within the inclusion criteria of the study.

This study is performed on CI users and normal hearing subjects. The study is considered to involve negligible risk and minimal burden. Benefits to the CI population at large may be a better and more efficient method of testing TFS performance. Given the small risk and high yield of this research it is an ethically justified study.