CN106714062A - BP-artificial-neural-network-based intelligent matching algorithm for digital hearing aid - Google Patents
BP-artificial-neural-network-based intelligent matching algorithm for digital hearing aid Download PDFInfo
- Publication number
- CN106714062A CN106714062A CN201611087426.0A CN201611087426A CN106714062A CN 106714062 A CN106714062 A CN 106714062A CN 201611087426 A CN201611087426 A CN 201611087426A CN 106714062 A CN106714062 A CN 106714062A
- Authority
- CN
- China
- Prior art keywords
- neural network
- artificial neural
- network
- digital hearing
- fitting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 71
- 238000012549 training Methods 0.000 claims abstract description 22
- 230000002068 genetic effect Effects 0.000 claims abstract description 17
- 230000006835 compression Effects 0.000 claims abstract description 16
- 238000007906 compression Methods 0.000 claims abstract description 16
- 230000003595 spectral effect Effects 0.000 claims abstract description 9
- 210000003127 knee Anatomy 0.000 claims abstract description 5
- 238000000034 method Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 9
- 238000013459 approach Methods 0.000 claims description 3
- 230000035772 mutation Effects 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 description 5
- 206010011878 Deafness Diseases 0.000 description 4
- 208000016354 hearing loss disease Diseases 0.000 description 4
- 206010010280 Conductive deafness Diseases 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 208000023563 conductive hearing loss disease Diseases 0.000 description 3
- 231100000895 deafness Toxicity 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 206010011891 Deafness neurosensory Diseases 0.000 description 1
- 208000009966 Sensorineural Hearing Loss Diseases 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000012076 audiometry Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000012074 hearing test Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 208000023573 sensorineural hearing loss disease Diseases 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
- H04R25/00—Deaf-aid sets, i.e. electro-acoustic or electro-mechanical hearing aids; Electric tinnitus maskers providing an auditory perception
- H04R25/50—Customised settings for obtaining desired overall acoustical characteristics
- H04R25/505—Customised settings for obtaining desired overall acoustical characteristics using digital signal processing
- H04R25/507—Customised settings for obtaining desired overall acoustical characteristics using digital signal processing implemented by neural network or fuzzy logic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
- H04R2225/00—Details of deaf aids covered by H04R25/00, not provided for in any of its subgroups
- H04R2225/43—Signal processing in hearing aids to enhance the speech intelligibility
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Automation & Control Theory (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Neurosurgery (AREA)
- Otolaryngology (AREA)
- Acoustics & Sound (AREA)
- Signal Processing (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
本发明公开了一种基于BP人工神经网络的数字助听器智能验配算法，该算法基于BP人工神经网络，通过大量的训练数据对网络进行训练，得出一个令人满意的成熟网络，并利用自行建立的公式模型对网络进行修正，从而得到成熟的智能验配算法；本发明利用遗传算法原理对BP人工神经网络的初始权值和阈值进行优化，并利用现有的听力图和频谱增益响应作为训练数据对BP人工神经网络进行训练，同时通过验配公式模型对网络进行修正，得到成熟的BP人工神经网络来代替现有的验配处方公式，进而得到数字助听器的各通道增益、最大声输出、压缩率以及压缩拐点等参数。
The invention discloses an intelligent fitting algorithm for digital hearing aids based on BP artificial neural network. The established formula model corrects the network, thereby obtaining a mature intelligent fitting algorithm; the present invention uses the genetic algorithm principle to optimize the initial weight and threshold of the BP artificial neural network, and uses the existing audiogram and spectral gain response as The training data trains the BP artificial neural network, and at the same time corrects the network through the fitting formula model, and obtains a mature BP artificial neural network to replace the existing fitting prescription formula, and then obtains the gain of each channel and the maximum sound output of the digital hearing aid , compression ratio and compression knee point and other parameters.
Description
技术领域technical field
本发明属于数字助听器领域的验配算法，更具体的说，是涉及一种基于BP人工神经网络的数字助听器智能验配算法，该算法基于BP人工神经网络，通过大量的训练数据对网络进行训练，得出一个令人满意的成熟网络，并利用自行建立的公式模型对网络进行修正，从而得到成熟的智能验配算法。The invention belongs to a fitting algorithm in the field of digital hearing aids, and more specifically, relates to an intelligent fitting algorithm for digital hearing aids based on a BP artificial neural network. The algorithm is based on a BP artificial neural network and trains the network through a large amount of training data. , get a satisfactory mature network, and use the self-established formula model to correct the network, so as to obtain a mature intelligent fitting algorithm.
背景技术Background technique
目前，中国人口老龄化问题日趋严重，到2030年中国老年人口比例将接近30％，耳聋是老年人的常见病，随着老年人口的增加，患有耳聋的老年人口的数量日益增多，人们对于助听器的需求与日俱增。At present, the problem of population aging in China is becoming more and more serious. By 2030, the proportion of the elderly population in China will be close to 30%. Deafness is a common disease of the elderly. With the increase of the elderly population, the number of elderly people suffering from deafness is increasing. The demand for hearing aids is increasing day by day.
数字助听器的诞生为助听器的发展奠定了坚实的基础，无论是语音信号处理的准确性还是应用的便捷性都有了长足的进步，这些性能的优化主要取决于动态范围压缩器，而动态压缩器是如何工作的主要取决于验配公式所得到的增益参数。数字助听器的验配公式有很多，主要分为以听阈为基础的和以响度为基础的，其中以听阈为基础的又分为线性的和非线性的两种。对于以响度为基础的主要有LGOB，线性的主要有NAL、DSL等，非线性的主要包括NAL-NL1、FIG6、DSL(i/o)等，目前最广泛应用的是POGO、NAL、DSL这三种。The birth of digital hearing aids has laid a solid foundation for the development of hearing aids. Both the accuracy of speech signal processing and the convenience of applications have made great progress. The optimization of these performances mainly depends on the dynamic range compressor, and the dynamic compressor How this works depends largely on the gain parameters derived from the fitting formula. There are many fitting formulas for digital hearing aids, which are mainly divided into those based on hearing threshold and those based on loudness. Among them, those based on hearing threshold are divided into two types: linear and nonlinear. The loudness-based ones mainly include LGOB, the linear ones mainly include NAL, DSL, etc., and the nonlinear ones mainly include NAL-NL1, FIG6, DSL (i/o), etc. Currently, the most widely used ones are POGO, NAL, and DSL. three kinds.
不同的验配公式对于同一个患者所得出的结果是不相同的，验配达到的效果也各有利弊。POGO公式是一个简单的半增益方法，即取得患者听阈的一半在加上经验常数得到通道增益，这种验配公式主要是对响度进行了调整，对于语言的可懂度的要求较少。NAL的发展经历了四代，现在的NAL-NL2已经非常成熟，尤其对于中等程度的耳聋患者的选配具有重要的实用价值。DSL验配公式主要适用于儿童。这几种常用的验配公式虽各有优点，但是对于听力损伤严重的耳聋患者效果较差，并且由于不同的患者需要选择不同的验配处方公式以达到最优的验配效果，这给验配工作带来了极大的不便。Different fitting formulas give different results to the same patient, and each fitting has its own advantages and disadvantages. The POGO formula is a simple half-gain method, that is, half of the patient’s hearing threshold is obtained and an empirical constant is added to obtain the channel gain. This fitting formula mainly adjusts the loudness, and has less requirements for language intelligibility. The development of NAL has gone through four generations, and the current NAL-NL2 is very mature, especially for the matching of patients with moderate deafness, which has important practical value. The DSL fitting formula is mainly applicable to children. Although these commonly used fitting formulas have their own advantages, they are less effective for deaf patients with severe hearing impairment, and because different patients need to choose different fitting prescription formulas to achieve the optimal fitting effect, this is a problem for the tester. The matching work has brought great inconvenience.
验配师希望有一种简单的验配方案，能够根据患者的听力图得出一个十分准确的验配参数，并且会随着验配案例的增多逐步完善，从而达到一个令人满意的验配效果。The fitter hopes to have a simple fitting plan, which can obtain a very accurate fitting parameter based on the patient's audiogram, and will gradually improve with the increase of fitting cases, so as to achieve a satisfactory fitting effect .
发明内容Contents of the invention
本发明的目的是为了克服现有技术中的不足，提供一种基于BP人工神经网络的数字助听器智能验配算法，该算法基于BP人工神经网络，通过大量的训练数据对网络进行训练，得出一个令人满意的成熟网络，并利用自行建立的公式模型对网络进行修正，从而得到成熟的智能验配算法；本发明利用遗传算法原理对BP人工神经网络的初始权值和阈值进行优化，并利用现有的听力图和频谱增益响应作为训练数据对BP人工神经网络进行训练，同时通过验配公式模型对网络进行修正，得到成熟的BP人工神经网络来代替现有的验配处方公式，进而得到数字助听器的各通道增益、最大声输出、压缩率以及压缩拐点等参数。The purpose of the present invention is to overcome the deficiencies in the prior art, provide a kind of digital hearing aid intelligent fitting algorithm based on BP artificial neural network, this algorithm is based on BP artificial neural network, train the network through a large amount of training data, draw A satisfactory mature network, and use the self-established formula model to correct the network, so as to obtain a mature intelligent fitting algorithm; the invention uses the principle of genetic algorithm to optimize the initial weight and threshold of the BP artificial neural network, and Use the existing audiogram and spectral gain response as training data to train the BP artificial neural network, and at the same time correct the network through the fitting formula model, and obtain a mature BP artificial neural network to replace the existing fitting prescription formula, and then Get the parameters of each channel gain, maximum sound output, compression ratio and compression knee point of the digital hearing aid.
本发明的目的是通过以下技术方案实现的：The purpose of the present invention is achieved through the following technical solutions:
一种基于BP人工神经网络的数字助听器智能验配算法，利用现有的听力图和频谱增益响应作为训练数据去训练BP人工神经网络，通过训练后的网络来估计预测新的患者，具体包括以下步骤：An intelligent fitting algorithm for digital hearing aids based on BP artificial neural network, which uses the existing audiogram and spectral gain response as training data to train BP artificial neural network, and estimates and predicts new patients through the trained network, including the following step:
(1)构建BP人工神经网络，包括输入层、隐含层和输出层；(1) Construct BP artificial neural network, including input layer, hidden layer and output layer;
(2)利用遗传算法对步骤(1)建立的BP人工神经网络的初始权值和初始阈值进行优化；(2) utilize genetic algorithm to optimize the initial weight and the initial threshold of the BP artificial neural network that step (1) establishes;
(3)对步骤(2)中经过优化后的BP人工神经网络进行训练，训练数据采用真实患者的案例，通过对BP人工神经网络的不断训练，使其逐步成熟，最终得到一个相对成熟的基于BP人工神经网络的数字助听器验配算法；(3) Train the optimized BP artificial neural network in step (2). The training data adopts the cases of real patients. Through continuous training of the BP artificial neural network, it gradually matures, and finally a relatively mature model based on BP artificial neural network digital hearing aid fitting algorithm;
(4)搭建用于对经过步骤(3)所述验配算法得到的增益输出进行优化的公式模型；(4) building a formula model for optimizing the gain output obtained by the fitting algorithm described in step (3);
(5)利用步骤(4)建立的所述公式模型对所述验配算法通过加权方式进行修正，以得到增益输出，随着后续训练数据的不断增多，加权系数会逐渐接近1，最终得到成熟完善的基于BP人工神经网络的数字助听器智能验配算法。(5) Use the formula model established in step (4) to correct the fitting algorithm by weighting to obtain gain output. With the continuous increase of subsequent training data, the weighting coefficient will gradually approach 1, and finally mature Perfect intelligent fitting algorithm for digital hearing aids based on BP artificial neural network.
步骤(1)中所述BP人工神经网络参数包括输入层和隐含层之间的权值Wij以及隐含层和输出层之间的权值Wjk，还包括隐含层和输出层的阈值。The BP artificial neural network parameters described in step (1) include the weight W ij between the input layer and the hidden layer and the weight W jk between the hidden layer and the output layer, and also include the weight W jk between the hidden layer and the output layer. threshold.
所述输入层、隐含层和输出层的节点数分别为10、20和30，按照频谱分为10个通道，即频率在250、500、750、1000、1500、2000、3000、4000、6000和8000Hz的目标为40dB的增益、压缩比和压缩拐点。The number of nodes in the input layer, the hidden layer and the output layer are 10, 20 and 30 respectively, and are divided into 10 channels according to the frequency spectrum, that is, the frequencies are 250, 500, 750, 1000, 1500, 2000, 3000, 4000, 6000 and 8000Hz target 40dB of gain, compression ratio and compression knee.
步骤(2)中所述优化流程如下：a.设定遗传算法的种群规模和迭代次数；b.将BP人工神经网络的初始权值和阈值进行GA编码，通过患者的听力图与频谱增益响应得到每个个体的适应度；c.最终经过遗传算法的选择操作、交叉操作和变异操作对初始权值和阈值进行种群适应度的计算，最终达到对网络阈值和权值的寻优。The optimization process described in step (2) is as follows: a. Set the population size and the number of iterations of the genetic algorithm; b. Perform GA encoding on the initial weights and thresholds of the BP artificial neural network, and pass the patient's audiogram and spectral gain response Get the fitness of each individual; c. Finally, through the selection operation, crossover operation and mutation operation of the genetic algorithm, the initial weight and threshold are calculated for the population fitness, and finally the optimization of the network threshold and weight is achieved.
与现有技术相比，本发明的技术方案所带来的有益效果是：Compared with the prior art, the beneficial effects brought by the technical solution of the present invention are:
1.为了更加准确的确定神经网络的初始权值和阈值，本发明利用遗传算法对神经网络进行优化，该优化算法已经非常成熟，遗传算法遵循“数竞人择，优者生存”的原则完成对网络初始权值和阈值的寻优；BP人工神经网络经遗传算法优化后，相对于原始的BP人工神经网络具备了先天的优势，通过将大量验配案例中的听力图和频谱增益响应对网络进行训练，随着训练数据的不断增加，基于BP人工神经网络的数字助听器智能验配算法也将越来越成熟。1. In order to determine the initial weight and threshold of the neural network more accurately, the present invention uses a genetic algorithm to optimize the neural network. The optimization algorithm is very mature, and the genetic algorithm is completed following the principle of "the number of candidates is selected, and the best survives". Optimizing the initial weights and thresholds of the network; the BP artificial neural network has inherent advantages compared to the original BP artificial neural network after being optimized by the genetic algorithm. The network is trained. With the continuous increase of training data, the intelligent fitting algorithm of digital hearing aids based on BP artificial neural network will become more and more mature.
2.为了使得验配的结果更加准确，本发明还利用自己建立的公式模型对BP人工神经网络进行了修正，通过两者的加权得出各通道增益，最终得到一个成熟的数字助听器智能验配算法。2. In order to make the fitting results more accurate, the present invention also uses the formula model established by itself to correct the BP artificial neural network, obtains the gain of each channel through the weighting of the two, and finally obtains a mature intelligent fitting of digital hearing aids algorithm.
3.在验配阶段，验配师只需要将听力测试的结果作为输入数据经本发明智能验配算法计算便可以得出令人满意的各通道增益、压缩比以及压缩拐点等参数，因为基于BP人工神经网络的数字助听器算法是经过大量验配案例训练得到的成熟网络，其输出的结果必然会接近患者真实的验配效果。3. In the fitting stage, the fitter only needs to use the results of the hearing test as input data to obtain satisfactory parameters such as the gain of each channel, the compression ratio and the compression inflection point through the calculation of the intelligent fitting algorithm of the present invention, because based on The digital hearing aid algorithm of BP artificial neural network is a mature network obtained through training of a large number of fitting cases, and the output result will inevitably be close to the real fitting effect of the patient.
4.本发明所涉及到的算法模型已经相对成熟，相比于其他的验配处方公式，基于BP人工神经网络的数字助听器智能验配算法所覆盖的范围更加广泛，并且由于该算法模型是经过真实验配案例所得到的成熟的算法模型，其输出的结果也将在一定程度上优于其他验配处方公式。本发明所涉及到的算法适合于数字助听器验配过程中代替现有的验配处方公式，必将为验配工作提供一个更加准确便捷的验配方案。4. The algorithm model involved in the present invention is relatively mature. Compared with other fitting prescription formulas, the intelligent fitting algorithm of digital hearing aids based on BP artificial neural network covers a wider range, and because the algorithm model is passed through The mature algorithm model obtained from the real test case will also output results that are better than other prescription formulas to a certain extent. The algorithm involved in the invention is suitable for replacing the existing fitting prescription formula in the digital hearing aid fitting process, and will surely provide a more accurate and convenient fitting solution for the fitting work.
附图说明Description of drawings
图1是本发明所用到的BP人工神经网络的拓扑结构示意图。Fig. 1 is a schematic diagram of the topology of the BP artificial neural network used in the present invention.
图2是本发明利用遗传算法优化BP人工神经网络并进行训练的具体流程图。Fig. 2 is the concrete flow chart of the present invention utilizes genetic algorithm to optimize BP artificial neural network and train.
图3是本发明所用到的公式模型验配方案流程图。Fig. 3 is a flow chart of the formula model fitting scheme used in the present invention.
图4是本发明所用到的公式验配原则与BP神经网络智能算法加权验配的流程图。Fig. 4 is a flowchart of the formula fitting principle and BP neural network intelligent algorithm weighted fitting used in the present invention.
具体实施方式detailed description
下面结合附图对本发明作进一步的描述：Below in conjunction with accompanying drawing, the present invention will be further described:
一种基于BP人工神经网络的数字助听器智能验配算法，实际上是利用现有的听力图和频谱增益响应作为训练数据去训练BP人工神经网络，通过训练后的网络来估计预测新的患者。进一步讲就是利用经过遗传算法原理优化的BP人工神经网络，通过现有的听力图和频谱增益响应作为训练数据对BP人工神经网络进行训练，同时利用验配公式的模型对训练后的网络进行修正，最终得到成熟的基于BP人工神经网络的数字助听器验配算法。具体过程步骤如下：An intelligent fitting algorithm for digital hearing aids based on BP artificial neural network actually uses the existing audiogram and spectral gain response as training data to train BP artificial neural network, and estimates and predicts new patients through the trained network. Further speaking, it is to use the BP artificial neural network optimized by the principle of genetic algorithm, and use the existing audiogram and spectral gain response as training data to train the BP artificial neural network, and at the same time use the model of the fitting formula to correct the trained network Finally, a mature digital hearing aid fitting algorithm based on BP artificial neural network is obtained. The specific process steps are as follows:
(1)构建BP人工神经网络，本实施例所使用的BP人工神经网络是一种多层次的前馈神经网络，即数据前向传递，误差后向传递。如图1所示，为BP人工神经网络的拓扑结构，BP人工神经网络是一种已经非常成熟的人工神经网络，主要结构包括输入层、隐含层和输出层。主要参数包括输入层和隐含层之间的权值Wij以及隐含层和输出层之间的权值Wjk，同时还包括隐含层和输出层的阈值。本发明中的验配算法模型规定BP人工神经网络的输入节点为10个，即频率在250、500、1000、2000、3000、4000、6000、8000Hz时的听阈以及性别(用1、2分别代表男和女)和经验(用1、2、3、4分别代表首次用户、短期用户、有经验用户和长期用户)。隐含层节点数为20，输出节点数为30，输出为10通道输出，即频率在250、500、750、1000、1500、2000、3000、4000、6000和8000Hz的目标为40dB的增益、压缩比和压缩拐点。用于训练此网络的训练数据均来自患者的真实验配案例，保证了网络的可靠性。(1) Construct a BP artificial neural network. The BP artificial neural network used in this embodiment is a multi-level feed-forward neural network, that is, data is transmitted forward and errors are transmitted backward. As shown in Figure 1, it is the topological structure of BP artificial neural network. BP artificial neural network is a very mature artificial neural network, and its main structure includes input layer, hidden layer and output layer. The main parameters include the weight W ij between the input layer and the hidden layer, the weight W jk between the hidden layer and the output layer, and also include the thresholds of the hidden layer and the output layer. The fitting algorithm model among the present invention stipulates that the input node of BP artificial neural network is 10, namely the hearing threshold and gender when the frequency is 250, 500, 1000, 2000, 3000, 4000, 6000, 8000 Hz (represented by 1, 2 respectively male and female) and experience (use 1, 2, 3, and 4 to represent first-time users, short-term users, experienced users, and long-term users, respectively). The number of hidden layer nodes is 20, the number of output nodes is 30, and the output is 10-channel output, that is, the target frequency is 40dB gain and compression at 250, 500, 750, 1000, 1500, 2000, 3000, 4000, 6000 and 8000Hz Ratio and compression knee. The training data used to train this network are all from the real experimental matching cases of patients, which ensures the reliability of the network.
(2)对步骤(1)建立的BP人工神经网络的初始权值和初始阈值进行优化；图2是进行优化并对网络进行训练的具体流程图，图1中已经具体阐述了本算法所利用的BP人工神经网络的具体参数，为了保证网络的最优性，本发明通过遗传算法对BP人工神经网络进行优化。遗传算法是一种对数值进行优化的算法，此处旨在为神经网络寻求最优的初始权值和阈值。本发明中规定遗传算法的种群规模为50个个体，迭代次数为1000次。具体优化步骤为将BP人工神经网络的初始权值和阈值进行GA编码，然后通过验配案例中的患者的听力图与频谱增益响应得到每个个体的适应度，然后经过遗传算法的选择操作、交叉操作和变异操作对初始权值和阈值进行种群适应度的计算，最终达到对网络阈值和权值的寻优。在训练过程中，BP神经网络根据输入的测听数据，依次通过隐含层节点的计算和输出节点的计算得出预测输出即各通道增益、压缩比和压缩拐点，预测输出通过与期望输出进行比对得出误差，误差通过网络进行反向传递更新网络中的阈值和权值，使得输出的结果逐渐逼近真实的输出结果。(2) optimize the initial weights and initial thresholds of the BP artificial neural network established in step (1); Fig. 2 is a specific flow chart for optimizing and training the network, which has been specifically described in Fig. 1. The specific parameters of the BP artificial neural network, in order to ensure the optimality of the network, the present invention optimizes the BP artificial neural network through a genetic algorithm. Genetic Algorithm is an algorithm for numerical optimization, here it aims to find the optimal initial weights and thresholds for the neural network. The present invention stipulates that the population size of the genetic algorithm is 50 individuals, and the number of iterations is 1000 times. The specific optimization steps are to GA code the initial weight and threshold of the BP artificial neural network, and then obtain the fitness of each individual through the audiogram and spectral gain response of the patient in the fitting case, and then through the selection operation of the genetic algorithm, The crossover operation and mutation operation calculate the population fitness for the initial weight and threshold, and finally achieve the optimization of the network threshold and weight. During the training process, the BP neural network obtains the predicted output through the calculation of the hidden layer node and the calculation of the output node according to the input audiometry data, that is, the gain of each channel, the compression ratio and the compression inflection point, and the predicted output is compared with the expected output. The error is obtained by comparison, and the error is transmitted backward through the network to update the threshold and weight in the network, so that the output result gradually approaches the real output result.
(3)对步骤(2)中经过优化后的BP人工神经网络进行训练，训练数据采用真实患者的案例，通过对BP人工神经网络的不断训练，使其步成熟，最终得到一个相对成熟的基于BP人工神经网络的数字助听器验配算法；(3) Train the optimized BP artificial neural network in step (2). The training data adopts the cases of real patients. Through continuous training of the BP artificial neural network, it becomes more mature, and finally a relatively mature model based on BP artificial neural network digital hearing aid fitting algorithm;
(4)搭建用于对经过步骤(3)所述验配算法得到的增益输出进行优化的公式模型；图3是搭建的公式模型验配方案的流程图。首先获取输入数据，主要有八通道听阈即频率在250、500、1000、2000、3000、4000、6000、8000Hz时的听阈和是否为传导性耳聋患者(是为1，不是为0)，然后取听阈的一半作为40dB输入时的增益并将采用平均值法将其改为十通道(频率在250、500、750、1000、1500、2000、3000、4000、6000、8000Hz的增益)，即将频率为500和1000Hz的增益取平均值作为750Hz的增益，将1000和2000Hz处的增益取平均作为1500Hz的增益，同时根据是否为传导性耳聋患者进行加6dB修改，最后通过压缩比(传导性耳聋患者取1.2，感音神经性耳聋患者取1.4)和压缩拐点对60dB和80dB输入时的增益进行计算，同时对输入80dB时250Hz处的增益进行减少6dB，在3000～4000Hz附近的增益减少3dB，同时根据输入听阈估测MPO。(4) Constructing a formula model for optimizing the gain output obtained through the fitting algorithm described in step (3); FIG. 3 is a flow chart of the fitting scheme of the formula model built. Firstly, the input data is obtained, mainly including eight-channel hearing thresholds, that is, the hearing thresholds at frequencies of 250, 500, 1000, 2000, 3000, 4000, 6000, and 8000 Hz and whether they are patients with conductive deafness (it is 1, not 0), and then take Half of the hearing threshold is used as the gain at 40dB input, and the average value method is used to change it to ten channels (the gain at frequencies of 250, 500, 750, 1000, 1500, 2000, 3000, 4000, 6000, 8000Hz), that is, the frequency is Take the average of the gains of 500 and 1000Hz as the gain of 750Hz, take the average of the gains at 1000 and 2000Hz as the gain of 1500Hz, and modify it by adding 6dB according to whether the patient is conductive deafness, and finally pass the compression ratio (the patient with conductive deafness takes 1.2, for patients with sensorineural deafness, take 1.4) and the compression inflection point to calculate the gain at 60dB and 80dB input. At the same time, the gain at 250Hz is reduced by 6dB when the input is 80dB, and the gain near 3000-4000Hz is reduced by 3dB. Enter the hearing threshold estimate MPO.
(5)利用步骤(4)建立的所述公式模型对所述验配算法通过加权方式进行修正，图4是本发明用到的公式验配原则与BP人工神经网络的智能算法进行加权验配的流程图，加权的目的是为了使得基于BP人工神经网络的智能算法的输出结果进行修正，使得输出的各通道增益更加逼近真实水平。具体方法是首先根据输入得到公式验配结果和智能验配算法验配结果，同时对生成的结果进行处理生成3X10的增益矩阵(频率在250、500、750、1000、1500、2000、3000、4000、6000、8000Hz时输入为40、60、80dB时的增益)，然后将公式验配得到的矩阵和智能算法得到的矩阵中的每一个增益进行加权得到总的增益矩阵，最后根据MPO对输出的结果进行限制，其中加权系数p是一个与网络相关的1X10的矩阵，即神经网络的倒数乘以一个常数k，随着网络的逐渐成熟k值会逐渐变大，直至p中的所有系数均变为1，此时标志网络的成熟，本发明基于BP人工神经网络的数字助听器智能验配算法就成熟且完善了。(5) Utilize the formula model established in step (4) to correct the fitting algorithm through a weighted manner, and Fig. 4 is a weighted fit using the formula fitting principle used in the present invention and the intelligent algorithm of the BP artificial neural network The purpose of weighting is to correct the output results of the intelligent algorithm based on BP artificial neural network, so that the output channel gain is closer to the real level. The specific method is to first obtain the fitting results of the formula and the fitting results of the intelligent fitting algorithm according to the input, and at the same time process the generated results to generate a 3X10 gain matrix (the frequency is 250, 500, 750, 1000, 1500, 2000, 3000, 4000 , 6000, 8000Hz when the input is the gain of 40, 60, 80dB), and then weight each gain in the matrix obtained by the formula fitting and the matrix obtained by the intelligent algorithm to obtain the total gain matrix, and finally according to MPO. As a result, the weighting coefficient p is a 1X10 matrix related to the network, that is, the reciprocal of the neural network is multiplied by a constant k. As the network matures, the value of k will gradually increase until all coefficients in p become equal. is 1, which indicates the maturity of the network, and the intelligent fitting algorithm for digital hearing aids based on the BP artificial neural network of the present invention is mature and perfect.
本发明并不限于上文描述的实施方式。以上对具体实施方式的描述旨在描述和说明本发明的技术方案，上述的具体实施方式仅仅是示意性的，并不是限制性的。在不脱离本发明宗旨和权利要求所保护的范围情况下，本领域的普通技术人员在本发明的启示下还可做出很多形式的具体变换，这些均属于本发明的保护范围之内。The present invention is not limited to the embodiments described above. The above description of the specific embodiments is intended to describe and illustrate the technical solution of the present invention, and the above specific embodiments are only illustrative and not restrictive. Without departing from the gist of the present invention and the scope of protection of the claims, those skilled in the art can also make many specific changes under the inspiration of the present invention, and these all belong to the protection scope of the present invention.
Claims (4)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611087426.0A CN106714062B (en) | 2016-11-30 | 2016-11-30 | An intelligent fitting method for digital hearing aids based on BP artificial neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611087426.0A CN106714062B (en) | 2016-11-30 | 2016-11-30 | An intelligent fitting method for digital hearing aids based on BP artificial neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106714062A true CN106714062A (en) | 2017-05-24 |
| CN106714062B CN106714062B (en) | 2020-02-18 |
Family
ID=58935394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611087426.0A Active CN106714062B (en) | 2016-11-30 | 2016-11-30 | An intelligent fitting method for digital hearing aids based on BP artificial neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106714062B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108703761A (en) * | 2018-06-11 | 2018-10-26 | 佛山博智医疗科技有限公司 | The test method of Auditory identification susceptibility |
| CN109151692A (en) * | 2018-07-13 | 2019-01-04 | 南京工程学院 | Hearing aid based on deep learning network tests method of completing the square certainly |
| CN109147808A (en) * | 2018-07-13 | 2019-01-04 | 南京工程学院 | A kind of Speech enhancement hearing-aid method |
| CN109714692A (en) * | 2018-12-26 | 2019-05-03 | 天津大学 | Noise reduction method based on personal data and artificial neural network |
| CN110166917A (en) * | 2018-02-16 | 2019-08-23 | 西万拓私人有限公司 | Method for adjusting the parameter of hearing system |
| CN110473567A (en) * | 2019-09-06 | 2019-11-19 | 上海又为智能科技有限公司 | Audio-frequency processing method, device and storage medium based on deep neural network |
| CN111491245A (en) * | 2020-03-13 | 2020-08-04 | 天津大学 | Digital hearing aid sound field identification algorithm based on cyclic neural network and hardware implementation method |
| CN111818436A (en) * | 2020-07-14 | 2020-10-23 | 无锡清耳话声科技有限公司 | Real ear analysis test system based on machine learning |
| CN112383870A (en) * | 2020-10-29 | 2021-02-19 | 惠州市锦好医疗科技股份有限公司 | Adaptive hearing parameter fitting method and device |
| CN112887885A (en) * | 2021-01-12 | 2021-06-01 | 天津大学 | Hearing aid fault automatic detection system and hearing aid system |
| CN116614757A (en) * | 2023-07-18 | 2023-08-18 | 江西斐耳科技有限公司 | Hearing aid fitting method and system based on deep learning |
| CN117241205A (en) * | 2023-10-18 | 2023-12-15 | 杭州惠耳听力技术设备有限公司 | Machine learning big data intelligent hearing aid verification method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1523219B1 (en) * | 2003-10-10 | 2008-08-20 | Siemens Audiologische Technik GmbH | Method for training and operating a hearing aid and corresponding hearing aid |
| CN104053112A (en) * | 2014-06-26 | 2014-09-17 | 南京工程学院 | A method for self-fitting hearing aids |
| CN105611477A (en) * | 2015-12-27 | 2016-05-25 | 北京工业大学 | Depth and breadth neural network combined speech enhancement algorithm of digital hearing aid |
| CN105722001A (en) * | 2014-12-23 | 2016-06-29 | 奥迪康有限公司 | Hearing Device Adapted For Estimating A Current Real Ear To Coupler Difference |
-
2016
- 2016-11-30 CN CN201611087426.0A patent/CN106714062B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1523219B1 (en) * | 2003-10-10 | 2008-08-20 | Siemens Audiologische Technik GmbH | Method for training and operating a hearing aid and corresponding hearing aid |
| CN104053112A (en) * | 2014-06-26 | 2014-09-17 | 南京工程学院 | A method for self-fitting hearing aids |
| CN105722001A (en) * | 2014-12-23 | 2016-06-29 | 奥迪康有限公司 | Hearing Device Adapted For Estimating A Current Real Ear To Coupler Difference |
| CN105611477A (en) * | 2015-12-27 | 2016-05-25 | 北京工业大学 | Depth and breadth neural network combined speech enhancement algorithm of digital hearing aid |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110166917A (en) * | 2018-02-16 | 2019-08-23 | 西万拓私人有限公司 | Method for adjusting the parameter of hearing system |
| CN108703761A (en) * | 2018-06-11 | 2018-10-26 | 佛山博智医疗科技有限公司 | The test method of Auditory identification susceptibility |
| CN109151692A (en) * | 2018-07-13 | 2019-01-04 | 南京工程学院 | Hearing aid based on deep learning network tests method of completing the square certainly |
| CN109147808A (en) * | 2018-07-13 | 2019-01-04 | 南京工程学院 | A kind of Speech enhancement hearing-aid method |
| CN109147808B (en) * | 2018-07-13 | 2022-10-21 | 南京工程学院 | A speech-enhancing hearing aid method |
| CN109714692A (en) * | 2018-12-26 | 2019-05-03 | 天津大学 | Noise reduction method based on personal data and artificial neural network |
| CN110473567A (en) * | 2019-09-06 | 2019-11-19 | 上海又为智能科技有限公司 | Audio-frequency processing method, device and storage medium based on deep neural network |
| CN111491245B (en) * | 2020-03-13 | 2022-03-04 | 天津大学 | Digital hearing aid sound field identification algorithm based on cyclic neural network and implementation method |
| CN111491245A (en) * | 2020-03-13 | 2020-08-04 | 天津大学 | Digital hearing aid sound field identification algorithm based on cyclic neural network and hardware implementation method |
| CN111818436A (en) * | 2020-07-14 | 2020-10-23 | 无锡清耳话声科技有限公司 | Real ear analysis test system based on machine learning |
| CN111818436B (en) * | 2020-07-14 | 2021-09-28 | 无锡清耳话声科技有限公司 | Real ear analysis test system based on machine learning |
| CN112383870A (en) * | 2020-10-29 | 2021-02-19 | 惠州市锦好医疗科技股份有限公司 | Adaptive hearing parameter fitting method and device |
| CN112383870B (en) * | 2020-10-29 | 2022-03-18 | 惠州市锦好医疗科技股份有限公司 | Adaptive hearing parameter fitting method and device |
| CN112887885A (en) * | 2021-01-12 | 2021-06-01 | 天津大学 | Hearing aid fault automatic detection system and hearing aid system |
| CN116614757A (en) * | 2023-07-18 | 2023-08-18 | 江西斐耳科技有限公司 | Hearing aid fitting method and system based on deep learning |
| CN116614757B (en) * | 2023-07-18 | 2023-09-26 | 江西斐耳科技有限公司 | Hearing aid fitting method and system based on deep learning |
| CN117241205A (en) * | 2023-10-18 | 2023-12-15 | 杭州惠耳听力技术设备有限公司 | Machine learning big data intelligent hearing aid verification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106714062B (en) | 2020-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106714062B (en) | An intelligent fitting method for digital hearing aids based on BP artificial neural network | |
| CN103079160A (en) | Automatic fitting digital hearing aid and use method thereof | |
| CN108364694A (en) | Airport Disease Warning Mechanism based on multi-data source big data and prevention and control system constituting method | |
| CN106786524B (en) | Parameter identification method of load model based on noise-like signal and improved differential evolution | |
| CN106991531A (en) | Power distribution network project construction effect Post-assessment Method based on BP neural network | |
| CN110457758B (en) | Method, device and system for predicting unstable phase of rock mass and storage medium | |
| CN106063295A (en) | A method of fitting a hearing aid system and a hearing aid fitting system | |
| CN110443447A (en) | A kind of method and system learning adjustment electric power system tide based on deeply | |
| CN104053112B (en) | A kind of audiphone tests method of completing the square certainly | |
| CN109714692A (en) | Noise reduction method based on personal data and artificial neural network | |
| CN112653980A (en) | Interactive self-checking and matching method for intelligent hearing aid | |
| CN110457757A (en) | Prediction method and device of rock mass instability stage based on multi-feature fusion | |
| CN106412841A (en) | Indoor location method based on DBSCAN algorithm | |
| CN106482738A (en) | Transfinited the indoor fingerprinting localization algorithm of learning machine based on online increment | |
| CN106618499B (en) | Fall detection equipment, fall detection method and device | |
| CN102769501A (en) | An Intelligent Identification Method of Upstream Channel Noise on Radio and Television Access Network | |
| CN111597713A (en) | Non-uniformity design flood calculation method considering flood season dispersion influence | |
| CN105178240B (en) | A kind of optimization method of drafting P type III distribution frequency curves | |
| CN110889250A (en) | A Damage Identification Method for Steel Truss Structures Based on Hybrid Meta-Heuristic Algorithm | |
| CN105892291A (en) | Robust control optimization method based on genetic algorithm | |
| CN111987720A (en) | Method for evaluating power receiving and power supply margin intervals of receiving-end power grid under constraint of multichannel quota | |
| CN111080462B (en) | Identification method of key users in social media based on non-uniform information dissemination characteristics | |
| CN107192898B (en) | Audible noise probability prediction method and system for direct current transmission line | |
| CN108197748B (en) | A People Flow Prediction Method Based on Thinking Evolution Algorithm | |
| CN108922616A (en) | A kind of hearing aid is quickly from testing method of completing the square |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20221129 Address after: 518172 516, building 4, Qidi Xiexin Science Park, Qingchun Road, Longcheng street, Longgang District, Shenzhen, Guangdong Province Patentee after: SHENZHEN EARTECH Co.,Ltd. Address before: 300072 Tianjin City, Nankai District Wei Jin Road No. 92 Patentee before: Tianjin University |
|
| TR01 | Transfer of patent right |