**2. Materials and Methods**

*2.1. Participants*

A total of 30 adults with at least 9 months of CI experience were recruited at the department of audiology and otoneurology of the Edouard Herriot University Hospital, Lyon, France (Figure 1).

**Figure 1.** Flow chart.

Eligible subjects were over 18 years old, suffered from bilateral profound hearing loss, and had had unilateral or bilateral CI for at least 9 months (range 1 to 26 years). All participants reported auditory difficulties in a noisy environment. The study protocol was approved by the local ethics committee (CPP Sud-Est IV 14/034 ID RCB 2014-A00345-42). Written informed consent was obtained from all patients.

The CI subjects were randomized into two groups using a computer-generated randomization list: an intervention group, which was called the study group (*n* = 15, 7 males, 8 females; mean age, 48 years, range 24 to 76 years) and an untrained group (control group) (*n* = 15, 8 males, 7 females; mean age 60 years, range 45 to 75 years). None of the subjects followed any other AT program during the study.

#### *2.2. Intervention*

With the aim of providing innovative and translational therapeutic methods in CI adults, a dedicated serious game was developed with the support of the French government ("Neurosyllabic R&D project"). The design and development of the serious game were based on previously published criteria for an effective AT protocol [10]. These criteria included ease of access (achievable at home and suitable for the elderly), interactivity, tasks of increasing complexity (to maintain the interest and attention of the subject), feedback, and the ability to record performances at any time.

A simple serious game scenario was developed in order to enable most subjects to easily identify with an avatar (Figure 2). Participants underwent a 5-week training program including 6 activities. The first 2 consisted in detecting and discriminating target sounds (animal calls, instruments, everyday noises, and words) in noise. These 2 activities were the only ones available during the 1st week. Then, 4 other games were introduced in the 2nd week: 1 consisted in target sound identification, and the last 3 were word-based games during which the subject had to either discriminate words, identify their syllables, or categorize them according to their semantic.

The auditory material included 240 noises, 22 instrument sounds, 100 animal calls, 3135 words, 665 logatomes, and 600 syllables, while the video material contained 1400 illustrative images. Among all the sounds, syllables, and words used, 30% were selected from a dedicated database created for the study, 40% were recorded by professional actors, and 30% (especially ambient background sounds) were purchased from a database on the Internet. In order for training to remain close to real-life conditions, while allowing a progressive increase in difficulty, for the first 2 games, subjects could choose from 4 types of ambient sounds each of which had a variable signal-to-noise ratio (SNR): white noise, continuous noises (sound of rain, wind, etc.), discontinuous noises (such as the auditory environments of everyday life), or babbling noises.

The game was automatically adapted in terms of difficulty. The volumes of the target sounds and the ambient sound (SNR) were adjusted according to 20 levels of difficulty. For levels 1 to 10, the target sounds were set at 100%, while the volume of the ambient sound increased from 0 to 90%. For levels 11 to 20, the ambient sound was set at 100%, while the volume of the target sounds decreased from 100% to 10%. The level of difficulty could either be set manually (in which case, each game had a fixed duration of 2 min) or adapted automatically by an algorithm (the game then stopped after 4 errors). In case of automatic management, the level of difficulty was set according to the previous games: it increased after each correct answer and decreased after each error. Adaptive changes in the difficulty level depended on 3 factors:

The probability of reaching a correct answer by chance (for instance, the increment in difficulty was lower if there was 1 correct answer among 2 than if there was 1 among 5).

Elapsed time: the more time passed, the greater the increment in difficulty and the smaller the decrement. This ensured that each game did not last too long.

The number of errors and correct answers that already occurred. A sequence of several mistakes without any correct answer since the beginning of the game meant that the initial level of difficulty was too high and therefore needed to be adjusted more quickly. Conversely, a faultless course led to a faster increase in difficulty.
