ABSTRACT
Background: Artificial intelligence (AI) chatbots, such as ChatGPT-4, have shown immense potential for application across various aspects of medicine, including medical education, clinical practice, and research. Objective: This study aimed to evaluate the performance of ChatGPT-4 in the 2023 Taiwan Audiologist Qualification Examination, thereby preliminarily exploring the potential utility of AI chatbots in the fields of audiology and hearing care services. Methods: ChatGPT-4 was tasked to provide answers and reasoning for the 2023 Taiwan Audiologist Qualification Examination. The examination encompassed six subjects: (1) basic auditory science, (2) behavioral audiology, (3) electrophysiological audiology, (4) principles and practice of hearing devices, (5) health and rehabilitation of the auditory and balance systems, and (6) auditory and speech communication disorders (including professional ethics). Each subject included 50 multiple-choice questions, with the exception of behavioral audiology, which had 49 questions, amounting to a total of 299 questions. Results: The correct answer rates across the 6 subjects were as follows: 88% for basic auditory science, 63% for behavioral audiology, 58% for electrophysiological audiology, 72% for principles and practice of hearing devices, 80% for health and rehabilitation of the auditory and balance systems, and 86% for auditory and speech communication disorders (including professional ethics). The overall accuracy rate for the 299 questions was 75%, which surpasses the examination's passing criteria of an average 60% accuracy rate across all subjects. A comprehensive review of ChatGPT-4's responses indicated that incorrect answers were predominantly due to information errors. Conclusions: ChatGPT-4 demonstrated a robust performance in the Taiwan Audiologist Qualification Examination, showcasing effective logical reasoning skills. Our results suggest that with enhanced information accuracy, ChatGPT-4's performance could be further improved. This study indicates significant potential for the application of AI chatbots in audiology and hearing care services.
Subject(s)
Artificial Intelligence , Audiologists , Audiology , Humans , Taiwan , Audiology/methods , Educational Measurement/methods , Male , Clinical Competence/standards , FemaleABSTRACT
This study aimed to preliminarily investigate the associations between performance on the integrated Digit-in-Noise Test (iDIN) and performance on measures of general cognition and working memory (WM). The study recruited 81 older adult hearing aid users between 60 and 95 years of age with bilateral moderate to severe hearing loss. The Chinese version of the Montreal Cognitive Assessment Basic (MoCA-BC) was used to screen older adults for mild cognitive impairment. Speech reception thresholds (SRTs) were measured using 2- to 5-digit sequences of the Mandarin iDIN. The differences in SRT between five-digit and two-digit sequences (SRT5-2), and between five-digit and three-digit sequences (SRT5-3), were used as indicators of memory performance. The results were compared to those from the Digit Span Test and Corsi Blocks Tapping Test, which evaluate WM and attention capacity. SRT5-2 and SRT5-3 demonstrated significant correlations with the three cognitive function tests (rs ranging from -.705 to -.528). Furthermore, SRT5-2 and SRT5-3 were significantly higher in participants who failed the MoCA-BC screening compared to those who passed. The findings show associations between performance on the iDIN and performance on memory tests. However, further validation and exploration are needed to fully establish its effectiveness and efficacy.
Subject(s)
Cognition , Cognitive Dysfunction , Hearing Aids , Memory, Short-Term , Humans , Aged , Female , Male , Middle Aged , Aged, 80 and over , Memory, Short-Term/physiology , Cognitive Dysfunction/diagnosis , Noise/adverse effects , Speech Perception/physiology , Speech Reception Threshold Test , Age Factors , Persons With Hearing Impairments/psychology , Persons With Hearing Impairments/rehabilitation , Hearing Loss/rehabilitation , Hearing Loss/diagnosis , Hearing Loss/psychology , Mental Status and Dementia Tests , Memory , Acoustic Stimulation , Predictive Value of Tests , Correction of Hearing Impairment/instrumentation , Auditory ThresholdABSTRACT
OBJECTIVE: This study aimed to develop a dual-task Mandarin Reading Span Test (RST) to assess verbal working memory related to speech perception in noise. DESIGN: The test material was developed taking into account psycholinguistic factors (i.e. sentence structure, number of syllables, word familiarity, and sentences plausibility), to achieve good test reliability and face validity. The relationship between the 28-sentence Mandarin RST and speech perception in noise was confirmed using three speech perception in noise measures containing varying levels of contextual and linguistic information. STUDY SAMPLE: The study comprised 42 young adults with normal hearing and 56 older adult who were hearing aid users with moderate to severe hearing loss. RESULTS: In older hearing aid users, the 28-sentence RST showed significant correlation with speech reception thresholds as measured by three Mandarin sentence in noise tests (rs or r = -.681 to -.419) but not with the 2-digit sequence Digit-in-Noise Test. CONCLUSION: The newly developed dual-task Mandarin RST, constructed with careful psycholinguistic consideration, demonstrates a significant relationship with sentence perception in noise. This suggests that the Mandarin RST could serve as a measure of verbal working memory.
ABSTRACT
OBJECTIVES: The study aimed to develop and validate the Mandarin digit-in-noise (DIN) test using four digit (i.e., two-, three-, four-, and five-digit) sequences. Test-retest reliability and criterion validity were evaluated. How the number of digits affected the results was examined. The research might lead to more informed choice of DIN tests for populations with specific cognitive needs such as memory impairment. DESIGN: The International Collegium of Rehabilitative Audiology guideline for developing the DIN was adapted to create test materials. The test-retest reliability and psychometric function of each digit sequence were determined among young normal-hearing adults. The criterion validity of each digit sequence was determined by comparing the measured performance of older adult hearing aid users with that obtained from two other well-established sentence-in-noise tests: the Mandarin hearing-in-noise test and the Mandarin Chinese matrix test. The relation between the speech reception thresholds (SRTs) of each digit sequence of the DIN test and working memory capacity measured using the digit span test and the reading span test were explored among older adult hearing aid users. Together, the study sample consisted of 54 young normal-hearing adults and 56 older adult hearing aid users. RESULTS: The slopes associated with the two-, three-, four-, and five-digit DIN test were 16.58, 18.79, 20.42, and 21.09 %/dB, respectively, and the mean SRTs were -11.11, -10.99, -10.56, and -10.02 dB SNR, respectively. Test-retest SRTs did not differ by more than 0.74 dB across all digit sequences, suggesting good test-retest reliability. Spearman rank-order correlation coefficients between SRTs obtained using the DIN across the four digit (i.e., two-, three-, four-, and five-digit) sequences and the two sentence-in-noise tests were uniformly high ( rs = 0.9) across all participants, when data from all participants were considered. Results from the digit span test and reading span test correlated significantly with the results of the five-digit sequences ( rs = -0.37 and -0.42, respectively) but not with the results of the two-, three-, and four-digit sequences among older hearing aid users. CONCLUSIONS: While the three-digit sequence was found to be appropriate for clinical use for assessment of auditory perception, the two-digit sequence could be used for hearing screening. The five-digit sequence could be difficult for older hearing aid users, and with its SRT related to working memory capacity, its use in the evaluation of speech perception should be investigated further. The Mandarin DIN test was found to be reliable, and the findings are in line with SRTs obtained using standardized sentence tests, suggesting good criterion validity.
Subject(s)
Hearing Aids , Speech Perception , Humans , Aged , Reproducibility of Results , Hearing Tests/methods , Noise , Language , Speech Reception Threshold TestABSTRACT
Machine-learning based approaches to speech enhancement have recently shown great promise for improving speech intelligibility for hearing-impaired listeners. Here, the performance of three machine-learning algorithms and one classical algorithm, Wiener filtering, was compared. Two algorithms based on neural networks were examined, one using a previously reported feature set and one using a feature set derived from an auditory model. The third machine-learning approach was a dictionary-based sparse-coding algorithm. Speech intelligibility and quality scores were obtained for participants with mild-to-moderate hearing impairments listening to sentences in speech-shaped noise and multi-talker babble following processing with the algorithms. Intelligibility and quality scores were significantly improved by each of the three machine-learning approaches, but not by the classical approach. The largest improvements for both speech intelligibility and quality were found by implementing a neural network using the feature set based on auditory modeling. Furthermore, neural network based techniques appeared more promising than dictionary-based, sparse coding in terms of performance and ease of implementation.