提出采用正弦模型改善患者高频听觉的非线性降频方法。正弦模型语音分解得到的幅度、频率和相位是算法三个主要的处理参数。为了避免谱失真,将语音频谱按倍频程划分为6个部分。最接近并低于患者门限频率的部分,只做幅度放大处理。按照不同频段对于语音理解度的贡献程度,将患者门限频率以上的频率段压缩并转移到患者的可听频段,并将对应相位信息变为最接近的对应低频相位。在本研究中,10个受试者进行了语音理解度测试。测试结果显示,经过训练后,患者的平均理解率至少提高45%。下一步的研究应增加受试者数量,并增加对患者的听损情况的详细分析,从而设计出更合理,更细致的降频助听算法。 This paper presents a nonlinear frequency reduction method using sinusoidal model to improve high-frequency hearing in patients. The amplitude, frequency and phase of the sinusoidal model speech decomposition are the three main processing parameters of the algorithm. In order to avoid spectral distortion, the speech frequency spectrum is divided into six parts by octave. The closest and lower than the threshold frequency of patients, only amplitude amplification. According to the contribution of different frequency bands to speech comprehension, the frequency band above the patient threshold frequency is compressed and transferred to the audible frequency band of the patient, and the corresponding phase information is changed to the closest corresponding low frequency phase. In this study, 10 subjects were tested for speech comprehension. Test results show that after training, the average rate of understanding of patients increased by at least 45%. The next study should increase the number of subjects and increase the detailed analysis of the patients hearing loss, in order to design a more reasonable and detailed frequency reduction helper algorithm.