OBJECTIVE Aimed to construct a corpus for a developing auditory-verbal communication training system tailored for the elderly and to grade the difficulty of sentence materials.The ultimate goal is to meet the diverse needs of trainees with varying hearing and speech capabilities.METHODS A training corpus,encompassing 3 627 sentences across 12 distinct topics,was systematically established.During the pre-experimental phase,12 adult subjects with normal hearing were recruited.They were tasked with conducting speech recognition rate tests on 100 randomly selected test sentences under four different signal-to-noise ratio(SNR)conditions.By plotting the performance-intensity(P-I)function curve,the SNR50 value,specifically-6.5 dB,was accurately determined.In the formal experiment,17 additional subjects were recruited.All sentences within the corpus were then subjected to speech recognition rate tests in a noisy environment under the established SNR50 condition.The average recognition rate was computed,and both the recognition rate-frequency histogram and the normal distribution curve were plotted.Leveraging the area beneath the normal curve,the segmentation points of the speech recognition rate were calculated,enabling the classification of sentences into three difficulty levels:easy,medium,and difficult.RESULTS The task of classifying the difficulty of sentences within the corpus was successfully accomplished.Specifically,there were 1 144 easy sentences,constituting 33%of the total;1 099 medium sentences,accounting for 32%;and 1 212 difficult sentences,representing 35%.CONCLUSION This research has successfully concluded the establishment of a corpus for the elderly's auditory-verbal communication ability training system and its associated difficulty grading.The obtained grading results are found to be in accordance with the principle of partitioning the difficulty gradient based on area.This study explores an effective method for grading the difficulty of auditory training sentences,ensuring that the training content matches the level of the trainer's competence.This not only enhances the training effectiveness but also improves the overall user experience.