研发并推广普适、普惠、普及且合用的助听器，是发展中国家助听器及其服务指南(第二版)的要求，也是解决我国老年性听力损失干预困局的利器。本课题为评估课题4所开发的国产高性能、免验配助听器产品：完善了中文言语测试材料与工具，研发了汉语普通话版开放式CMnBio语句测听材料及闭合式听音识图”类辅音识别材料；开发了适用于中老年人听力自评估的(PC端&移动设备端)言语测听系统并获得软件著作权；建立便捷的中文语境下助听临床评估框架；开展临床对照研究，完成了157例国产高性能助听器与国外同档次助听器的临床对比研究及71例传统验配模式与自主验配模式前瞻性随机对照单盲研究；此外，根据以上研究结果，撰写了国产高性能、免验配助听器的应用示范总体方案与实施细则。WHO(2004)announced the guidelines containing requirements and recommendation for developing countries in which hearing aids and services shall be appropriate,acceptable,affordable and available.It shall also be the approach to resolve the dilemma related with the low prevalence of hearing aid acquisition in Chinese older adults.To evaluate the high-performance,self-fitting hearing aid which were designed and product by Study Ⅳ,we improved Mandarin speech test materials,including a Mandarin Chinese open-set sentence material,CMnBio,and a closed-set“WIPI”(the world intelligibility by picture identification)consonant recognition material.Meanwhile,we developed the speech audiometry system(PC software&App)for elderly listening self-evaluation,whose copyright is obtained.Then,a comparative study on the performance of the high-performance,self-fitting hearing aid made in China and abroad(157 clinical trials),and a prospective-randomized study on the effect between routine-fitting and self-fitting hearing aids(71 clinical trials)are carried out.Based on the study conclusion,we drew out the program of demonstration and implementation of high-performance,self-fitting hearing aids application in China.

针对经典助听器听力补偿算法在噪声和干扰声等场景中性能下降的问题，本课题研究基于现代音频信号处理和人工智能技术的新型助听器语音处理算法。在通道自适应的宽动态范围压缩方面，提出了增益平稳调节的动态范围控制算法和自适应滤波器的助听器宽动态范围压缩算法；在基于人工智能技术的算法参数优化方面，提出了基于深度域自适应网络的迁移学习模型，融合时序卷积网络和残差块；在助听器自适应语音方向性合成技术方面，提出融合广义旁瓣对消和自适应陷波滤波器的助听器指向性算法、基于信噪比的数字助听器自适应指向性算法、基于回归器和密集连接卷积网络的声方位估计算法；在基于自适应环境识别的多类噪声抑制及反馈消除算法研究方面，提出了非线性回声抑制方法、基于互相关检测与时频掩蔽的助听器风噪声抑制算法、面向瞬态噪声抑制的语音增强方法，在深度神经网络方面，提出了基于混合空洞卷积网络和残差密集网络、面向麦克风阵列的通道注意力加权、基于特征维度自注意力的语音增强方法。本课题共发表了论文13篇，其中SCI收录9篇，申请专利13项，其中授权8项，研究成果降低了各种环境干扰对助听器使用的影响，实现了助听器感知语音质量的提升，提高了患者的言语交流能力和助听器使用舒适度，获得更好的助听效果，同时系统计算量小，实时性好，满足了数字助听器的低功耗高实时性设备的要求，提升国产数字化助听器等听力康复设备的技术水平。Aiming at the problem of performance degradation of classical hearing aid hearing compensation algorithms in scenarios such as background noise and interference signals,this project investigates new hearing aid speech processing algorithms based on modern speech signal processing and artificial intelligence technology.In terms of channel adaptive wide dynamic compression,a dynamic range control algorithm with smooth gain adjustment and an adaptive filter wide dynamic range compression algorithm for hearing aids are proposed;in terms of algorithmic parameter optimization based on artificial intelligence technology,a transfer learning model based on a deep domain adaptive networks fusing temporal convolutional networks and residual blocks is proposed;in terms of adaptive speech directionality synthesis for hearing aids,directionality algorithm integrating generalized sidelobe canceller and adaptive notch filter,a digital hearing aid adaptive directionality algorithm based on signal-to-noise ratio,and an sound source localization algorithm based on regressor and densely connected convolutional network are proposed;In terms of multi-class noise suppression and feedback cancellation algorithms based on adaptive environment recognition,nonlinear echo suppression methods,wind noise suppression algorithms for hearing aids based on cross correlation detection and time-frequency masking are proposed.Also,for deep neural network-based speech enhancement,speech enhancement methods based on hybrid dilated convolution networks and residual dense networks,microphone array-based channel attention weighting,and feature dimension self-attention are proposed.A total of 13 papers have been published,of which 9 are SCI-indexed,and 13 patents have been applied for,of which 8 are authorized.The research reduces the impact of various environmental interference on the use of hearing aids,achieve the improvement of the perceived speech quality of hearing aids,improve the patient's speech communication ability and the comfort of using hearing aids,and obtain better hearing aid effects.Also,the system has low computation complexity,which meets the requirements of low-power and high real-time equipment for digital hearing aids.This research improves the technical level of domestic digital hearing aids and other hearing rehabilitation equipment.

针对传统助听器在存在的问题，本课题主要研究高性能及免验配助听器的系统电路设计、算法优化与实现、通用声学结构设计等关键技术，攻克高性能助听器音频处理算法与芯片算力与功耗的矛盾，突破基于模式切换的免验配助听器个性化补偿效果不佳的问题，完成性能指标达到国际领先水平的高性能助听器及适用于广大听障人群自调佩戴的免验配助听器产品。_x000D_课题从关键元件选型、电路设计、声结构设计、算法移植、验配软件研发等方面开展研究完成了高性能全数字助听器和免验配全数字助听器的样机研发。_x000D_课题基于数字助听器专用信号处理平台进行了高性能全数字助听器和免验配全数字助听器的算法实现、样机研发以及验配软件开发。课题组对信号处理平台的特点、体系架构、固件库、指令集及各种资源进行了充分学习和调研，在此基础上开展相关工作。_x000D_根据高性能与免验配数字助听器的要求选择了能满足需求的传声器和受话器元件，对高性能数字助听器，特别选择了预匹配的全向传声器。综合考虑助听器芯片、助听器声腔结构等因素设计了助听器柔性电路板。针对助听器啸叫引起佩戴者极大不适的问题，设计了一种抑制数字助听器产生啸叫的声结构。针对高性能和免验配助听器，研制了PC端验配软件和验配软件手机APP,可通过HIPRO专用设备或者蓝牙调节助听器的音量、程序、增益、通道、附加功能等等，为用户提供远程验配，听力测试以及对应听力图的自主验配等。In response to the problems existing in traditional hearing aids,this project mainly focuses on key technologies such as system circuit design,algorithm optimization and implementation,and universal acoustic structure design for high-performance and exempt hearing aids.It aims to overcome the challenges of audio processing algorithms and chip computing power and power consumption in high-performance hearing aids,and break through the problem of poor personalized compensation effect of exempt hearing aids based on mode switching,High performance hearing aids that achieve internationally leading performance indicators and are suitable for self adjusting and wearing by a large number of hearing impaired individuals._x000D_The project has conducted research and completed the prototype development of high-performance all digital hearing aids and test free all digital hearing aids from the aspects of key component selection,circuit design,sound structure design,algorithm transplantation,and testing software development._x000D_The project is based on a dedicated signal processing platform for digital hearing aids,and the algorithm implementation,prototype development,and fitting software development of high-performance all digital hearing aids and non matching all digital hearing aids have been carried out.The research team has thoroughly studied and researched the characteristics,architecture,firmware library,instruction set,and various resources of the signal processing platform,and carried out relevant work on this basis._x000D_We have selected microphone and receiver components that can meet the requirements of high-performance and test free digital hearing aids.For high-performance digital hearing aids,we have particularly selected pre matched omnidirectional microphones.A flexible circuit board for hearing aids was designed taking into account factors such as the hearing aid chip and the structure of the hearing aid cavity.A sound structure is designed to suppress the squealing of digital hearing aids,which can cause great discomfort to the wearer.We have developed a PC based matching software and matching software mobile app for high-performance and OTC hearing aids.The software and app can adjust the volume,program,gain,channel,and additional functions of the hearing aid through HIPRO dedicated devices or Bluetooth,providing users with remote matching,hearing testing,and independent matching of corresponding hearing maps.

本年度进行了实现高性能全数字助听器和免验配全数字助听器的基础功能研发及研究低功耗数字信号处理器的并行计算方案和进行助听器算法的实现与优化。DSP芯片是高性数字助听器各种算法、功能实现的基础，针对数字助听器的特点及本课题的任务指标，选用Acod3100芯片作为平台，分析了Acod3100芯片的资源和特点。根据Acod3100芯片电气参数及任务指标，进行了高性能和免验配数字助听器关键元件传声器和受话器的分析和选型。为了提高高性能和免验配数字助听器的用户体验，以便提高国产数字助听器的用户接受度，制定了助听器辅助功能方案。This year,subject 4 focused on the basic functions of high-performance digital hearing aids and OTC digital hearing aids,as well as the parallel computing scheme,the implementation and optimization of hearing aid algorithms based on low-power digital signal processor.The DSP chip is used to implement a variety of algorithms and functions in high-performance digital hearing aids.The acod3100 chip is chosen as the platform based on the characteristics of digital hearing aids and the task indicators for this subject,and the resources and characteristics of the acod3100 chip are analyzed.The microphone and receiver,which are crucial components of high-performance and OTC digital hearing aids,are studied and selected based on the electrical properties of the acod3100 chip and the task indicators.The auxiliary function scheme of hearing aid is formulated to increase the user experience of high-performance and OTC digital hearing aids,as well as the acceptance of domestic hearing aids。

听力障碍严重影响老龄患者身心健康，中国的老龄化现状使得这一问题尤为突出,65 岁以上的老年人中约有三分之一患有听力障碍，研发适合基层推广和老人使用的智能听力康复辅具迫在眉睫，具有重要的社会意义。助听器作为听力障碍人群补偿听力功能、改善生存状况的最常用康复辅具，根据助听方式不同可分为气传导助听器和骨传导助听器。骨传导助听方式是通过颅骨振动将声能量传导至内耳，再由内耳解析为神经电信号从而刺激听觉神经形成声音感知过程。本年度课题组主要通过文献调研分析国内骨传导助听器应用现状，包括技术特点、产品标准以及其临床适应症，旨在推广骨传导助听器临床应用。Hearing aids are the most frequently used rehabilitation aids for people with hearing impairment to compensate their hearing function and improve their living conditions.It can be divided into air conduction hearing aid and bone conduction hearing aid in terms of sound transmission pathway.On the bone conduction,the sound signal vibrates the skull and is transmitted into the inner ear,and then it can be interpreted as a neural electrical signal to stimulate the auditory nerve until the sound perception process being formed.Based on literature research,the paper mainly analyzes the domestic status of bone conduction hearing aid application including technical characteristics,product standards and clinical indications,which Hearing aids are the most frequently used rehabilitation aids for people with hearing impairment to compensate their hearing function and improve their living conditions.It can be divided into air conduction hearing aid and bone conduction hearing aid in terms of sound transmission pathway.On the bone conduction,the sound signal vibrates the skull and is transmitted into the inner ear,and then it can be interpreted as a neural electrical signal to stimulate the auditory nerve until the sound perception process being formed.Based on literature research,the paper mainly analyzes the domestic status of bone conduction hearing aid application including technical characteristics,product standards and clinical indications,which aims to promote clinical application of bone conduction hearing aid.

本年度课题三对高性能助听器的核心算法及关键技术进行了深入研究。提出了通道自适应的数字助听器宽动态范围压缩算法，根据人耳的听觉特性及输入信号的强度信息确定滤波器组的中心频率向量及相关参数，结合听损患者的听力图进行通道合并，获得符合患者听损的自适应通道滤波器组，对每个通道的听力损失进行增益补偿，采用最少的通道数提供高质量适配患者听力图的补偿语音。提出了基于信噪比的数字助听器自适应指向性算法，将自适应滤波结构的噪声参考信号与系统的输出信号相结合，在降低计算复杂度的同时，提升了信干比。提出了一种基于双路麦克风信号的风噪声抑制算法，利用两路风噪声信号的互相关系数明显低于语音相关性系数的特点，对风噪声实现有效检测，在纯风噪情况下实现风噪抑制，在含风噪语音下，可以调节噪声的抑制深度和语音音质。提出了适用于声音场景分类的融合注意机制的集成卷积神经网络，将声谱图特征作为输入，并划分为不同的大小，利用卷积神经网络进行学习，同时引入全局注意层来计算全局特征相关性，在输出层计算不同分段级特征的权重进行融合，实验结果证明了该方法的有效性。提出了基于混合空洞卷积网络和基于残差密集网络的语音增强算法，引入混合空洞卷积和密集连接层，利用局部特征融合和局部残差学习挖掘了对数功率谱的二维信息，有效提升了增强语音的感知质量和可懂度。This year our group deeply investigated the core algorithms and key technologies of high-performance hearing aids.An adaptive-channel wide dynamic range compression algorithm is proposed.Multichannel filterbank decomposes signals in the loudness compensation algorithm of hearing aids.The center frequency and parameters of the filterbank are determined according to the human auditory characteristics and the input signal.Some adjacent channels merges into one channel based on the special audiogram.The sound is compensated in each channel.The propose algorithm reduces the computational complexity.An adaptive directivity algorithm based on signal-to-noise ratio is proposed.The signal-to-noise ratio is calculated and determines the update step of the adaptive filtering coefficient.The Signal to Interference Ratio of the output signal are improved,while reducing the computational complexity.A wind noise suppression algorithm based on two microphone signals is proposed.If the cross-correlation coefficient is lower than the set threshold,the signal is considered as wind noise signal,and the time-frequency masking technology is applied to suppress it.This method can effectively suppress wind noise in the case of wind noise.On the condition of speech with wind noise,the coefficients are adjusted to make a tradeoff between the noise suppression and speech quality.A convolutional neural network with attention mechanism is proposed for acoustic scene classification.Mel-spectrograms are divided into different sizes according to the frame coordinates as the feature of convolutional neural networks.In addition,the global attention layer is interspersed in the convolution network to calculate the global feature dependency,and the weights of different segment-level features are calculated in the output layer for fusion.Experimental results demonstrate the effectiveness of the proposed approach.A speech enhancement algorithms based on Hybrid Dilated Convolution Network(HDCN)and a residual dense network are proposed.HDC and dense connection layers are introduced.Local feature fusion and local residual learning combine the features of all network layers.The proposed methods effectively improve the perceived quality and intelligibility.

本年度中，课题二针对高性能验配助听器不适合基层推广的现状，围绕全数字助听器验配信息综合性数据库的构建、基于多参数拟合的个体听力损失建模、基于异构数据融合的助听器免验配方法和助听器交互式智能验配方法四个主要内容开展研究工作。为了更好的体现免验配助听器的免验配特性，提出一种基于助听器增益的个体听损建模方法，将不同听损的听障患者进行分类，并选择代表性听力信息来表征听损患者间的差异，从而避免繁琐的听力专家验配；为了能更好引导免验配助听器的参数优化，提出一种面向免验配助听器的参数自调节方法，根据耳蜗听损模型建立响度补偿模型，并基于声压级和语音可懂度构建一个对响度补偿效果的评价指标，并通过参数优化算法使响度补偿效果最优化。为了改善免验配助听器使用者的用户体验，改善算法性能，提出一种智能助听器交互式自验配方法，既充分考虑了利用患者的综合性信息和验配的历史信息，又基于智能算法设计模型模拟听力专家实现参数调整最优化。此外，为了有效的扩充数据集，本课题除了采集真实数据外，还从美国的全国健康和营养检查调查(NHANES)收集了大量的听力数据。本年度本课题共授权国家发明专利2项，申请发明专利2项，申请团体标准(个人听力信息码数据格式)1项，指导硕士毕业生1人。In this year,aiming at the current situation that high-performance hearing aid fitting is not suitable for grass-roots promotion,our group carried out research work around four main contents:the construction of comprehensive database of digital hearing aid fitting information,individual hearing loss modeling based on multi parameter fitting,the key technology of OTC hearing aid based on heterogeneous data fusion and hearing aid interactive intelligent fitting method.In order to better design OTC hearing aids,an individual hearing loss modeling method based on hearing aid gain is proposed.Hearing impaired patients with different hearing losses are classified,and representative hearing information are selected to represent the differences of hearing impaired patients,so as to avoid complicated hearing expert fitting;In order to better guide the parameter optimization of OTC hearing aids,a parameter self-adjusting method is proposed.The loudness compensation model is established according to the cochlear hearing loss model,and an evaluation index of loudness compensation is constructed based on sound pressure level and speech intelligibility,so as to optimize the loudness compensation effect by using the parameter optimization algorithm.In order to improve the user experience of hearing aid users and improve the performance of the algorithm,an interactive self-fitting method of intelligent hearing aid is proposed,which not only takes full account of the comprehensive information of patients and the historical information of fitting,but also simulates hearing experts to adjust and optimize parameters based on the intelligent model.In addition,in order to effectively expand the data set,in addition to collecting real data,our group also collected a large number of hearing data from the National Health and Nutrition Examination Survey(NHANES)in the United States.In this year,the subject has authorized 2 national invention patents,applied for 2 invention patents,applied for 1 group standard(personal hearing information code data format),and guided 1 master's graduate.

高性能、免验配助听器的研发目标是突破制约我国老年人助听器使用的二大瓶颈——性价比低、基层验配能力不足。本课题将首先构建全新的中文语境下助听器效果主客观评估框架，立足于课题4开发的国产高性能、免验配助听器产品，评估其与国内外同档次、传统验配模式助听产品的性能优劣，并制定智能助听器应用示范总体方案与实施细则。1)对照分析了通行的老年听力残障调查表(HHIE)及助听器日常生活使用满意度调查表(SADL)的中英文版本，提出修订意见。同时，也拟采用跨文化翻译法，参考BHI(Better Hearing Institute)Quick Hearing Check(2010)或Hearing Screening Inventory(HIS,1992)对老年人采用问卷方式进行听力自评估，以分析不同年龄段的老年人听力损失知晓度。现已初步制订了涉及老人互联网应用经验的调查问卷《手机App使用行为调研问卷》。2)为编撰新的中文言语测听语料，进行了文献回顾，发表了一篇综述《数字三联音测试规范及其在老年听力筛查中的应用》3)开展了有关免验配的OTC助听器与传统验配模式助听器的临床对照的预研。招募20名70岁以下老年轻~中重度听障老人，采用前瞻性的单盲、随机、对照实验，已完成数据收集，正在分析整理、撰写文章。The revisions were made in contrast to the Chinese and English common versions of the Hearing Handicap Inventory for the Elderly(HHIE)and the Satisfaction with Amplification in Daily Life Scale(SADL).It is also proposed to use cross-cultural translation,using BHI(Better Hearing Institute)Quick Hearing Check(2010)or Hearing Screening Inventory(HIS,1992)to conduct a self-assessment of hearing in older people using questionnaires to analyze hearing loss awareness among the older people of different ages.A questionnaire on the experience of Internet applications for the elderly,The Mobile App Usage Behavior Survey,has been initially developed.we are carrying out the literature review in order to compile a new set of Chinese speech audiometric materials.One of review on the Standardization of the Digital Triplet Test(DTT)and Its Application in Geriatrics,was published in 2020.A pilot study on outcome comparison of self-fitting OTC hearing aids and audiologist-fitting traditional hearing aids was carried out.Twenty moderate-to-severe hearing-impaired participants aged 40~70 were recruited.The design of this study was a single-site,prospective,single-blind controlled randomized trial with two parallel branches:(a)audiology best practices(AB),(b)consumer decides OTC model(CD).We have completed data collection,is statistical analyzing the results in order to publish the paper soon.

针对传统助听器在存在的问题，本课题主要研究高性能及免验配助听器的系统电路设计、算法优化与实现、通用声学结构设计等关键技术，攻克高性能助听器音频处理算法与芯片算力与功耗的矛盾，突破基于模式切换的免验配助听器个性化补偿效果不佳的问题，完成性能指标达到国际领先水平的高性能助听器及适用于广大听障人群自调佩戴的免验配助听器产品。本年度课题四进行了高性能全数字助听器和免验配全数字助听器的软硬件方案设计。DSP芯片是高性能及免验配数字助听器各种算法、功能实现的基础，针对数字助听器的特点及本课题的任务指标，选用Onsemi公司的Ezairo7160芯片作为算法实现平台，分析了Ezairo7160芯片的资源和特点。根据Ezairo7160芯片电气参数及任务指标，进行了高性能和免验配数字助听器关键元件传声器和受话器的分析和选型。为了提高高性能和免验配数字助听器的用户体验，以便提高国产数字助听器的用户接受度，制定了助听器辅助功能方案。In view of the existing problems of traditional hearing aids,this project mainly studies key technologies such as system circuit design,algorithm optimization and implementation,general acoustic structure design of high-performance and OTC hearing aids.It overcomes the contradiction between high-performance hearing aid algorithms and power consumption,and resovles the problem of poor personalized compensation for OTC hearing aids based on mode switching.It aims to accomplish the development of high-performance hearing aids reaching international advanced level,and OTC hearing aid suitable for self-adjusting of hearing-impaired people.This year,the project designs the software and hardware scheme of high-performance hearing aids and OTC hearing aids.Various algorithms and functions are implemented on DSP chip,which is the platform of high-performance and OTC hearing aids.According to the characteristics of digital hearing aids and the task target of this subject,the Ezairo7160 chip of Onsemi is selected as the algorithm realization platform.In accordance with the electroacoustics parameters of the subject and electrical parameters of the Ezairo7160 chip,the types of microphones and receivers are analyzed and selected,meanwhile,the circuit and structure of high-performance hearing aids are designed.In order to improve the user experience of high-performance and OTC digital hearing aids,the scheme of auxiliary functions and fitting software are formulated.

本年度课题三对高性能助听器的核心算法及关键技术进行了输入研究。针对数字助听器噪声抑制的问题，构建基于深度神经网络的语音增强模型，提高语音增强的鲁棒性，适用于助听器系统。研究了增益平稳调节的助听器动态范围控制算法，解决经典算法存在增益波动较大，导致语音信号音质受损的问题，能够不失真地对助听器声音信号的幅度进行控制，提高输出语音信号音质。针对数字助听器系统的回声和啸叫问题，研究了一种非线性回声抑制方法，在频域进行回声抑制，对本地语音的畸变小，可以最大程度保证输出语音的音质，计算量小，实时性好，满足数字助听器的低功耗高实时性要求。针对助听器的自适应指向性技术，研究了基于自适应滤波结构的双传声器指向性系统。针对传声器间距小的情况，采用级联积分梳状滤波器进行插值滤波的方案实现分数延迟。基于重叠存储法设计频域分块LMS算法，在每个自适应周期的计算量少于传统的LMS算法。提出了基于深度卷积网络的语音定位算法，引入了密集连接结构和Soft-argmax回归结构，提升定位的精确性和鲁棒性。在方向性语音增强研究方面，研究了一种面向麦克风阵列的通道注意力加权的语音增强方法。采用U-NET网络对语音频谱特征进行处理，并对通道维度采用注意力机制加权操作，在深层结构模拟波束形成对特征进行加权，提高语音增强的性能。This year our group deeply investigated the core algorithms and key technologies of high-performance hearing aids.According to the noise reduction in digital hearing aids,a speech enhancement model is built based on Deep Neural Network(DNN)to improve the robustness.The proposed algorithm has the prospect for hearing aids.A dynamic range control algorithm is studied.In the algorithm the gain is adjusted smoothly to reduce the excessive fluctuation and speech distortion.Through the proposed algorithm the amplitude of the speech is controlled and the quality of the output speech is improved.Aiming at the problems of the echo and howling in digital hearing aids,a nonlinear echo cancellation method is studied.The echo is estimated and inhibited in the frequency domain to preserve the speech quality.The proposed algorithm has moderate computational complexity and good real-time performance to meet the requirements of the low power consumption devices,such as digital hearing aids.With regard to adaptive directivity technology of hearing aids,a dual microphone directional system based on adaptive filtering structure is studied.Cascade Integrator Comb(CIC)filter is used to interpolate and filter to achieve fractional delay in the case of small microphone spacing.LMS algorithm based on overlapping storage is designed,and the computational amount of each adaptive period is less than that of traditional LMS algorithm.The training and network structure of the convolutional neural network(CNN)are analyzed,and a speech localization algorithm based on CNN is proposed.This algorithm introduces a dense connection structure.Also,Soft-argmax regressor structure is also utilized,which improves the accuracy and robust of speech localization.In the research of directional speech enhancement,a method of speech enhancement based on channel attention weighting for microphone array is studied.U-NET network is used to process the voice spectrum characteristics,and the channel dimension is weighted by the attention mechanism.For each layer of encoder and decoder,the attention is weighted in the channel dimension,and the beam-former is simulated in the deep structure to weight the features,which can improve the performance of speech enhancement.