A Non-Intrusive Speech Quality Evaluation Method Based on the Audiogram andWeighted Frequency Information for Hearing Aid

摘要

For a long time, the compensation effect of hearing aidis mainly evaluated subjectively, and there are fewer studies of objectiveevaluation. Furthermore, a pure speech signal is generally required as areference in the existing objective evaluation methods, which restricts thepracticality in a real-world environment. Therefore, this paper presents anon-intrusive speech quality evaluation method for hearing aid, which combinesthe audiogram and weighted frequency information. The proposedmodel mainly includes an audiogram information extraction network, afrequency information extraction network, and a quality score mapping network.The audiogram is the input of the audiogram information extractionnetwork, which helps the system capture the information related to hearingloss. In addition, the low-frequency bands of speech contain loudness informationand the medium and high-frequency components contribute tosemantic comprehension. The information of two frequency bands is inputto the frequency information extraction network to obtain time-frequencyinformation. When obtaining the high-level features of different frequencybands and audiograms, they are fused into two groups of tensors that distinguishthe information of different frequency bands and used as the inputof the attention layer to calculate the corresponding weight distribution. Finally,a dense layer is employed to predict the score of speech quality. Theexperimental results show that it is reasonable to combine the audiogramand the weight of the information from two frequency bands, which caneffectively realize the evaluation of the speech quality of the hearing aid.