Gain adjustment at tinnitus pitch to manage both tinnitus and speech perception in noise

To investigate how much gain variation is required from prescription to effect tinnitus percept,and if this revised prescription affects speech recognition.Twenty participants who experienced catastrophic tinnitus even after fitted with hearing aid were included.Participants were grouped based on their tinnitus pitch and the prescriptive formula used to fit hearing aid.They were evaluated for handicap from tinnitus using Tinnitus Handicap Inventory(THI).Hearing aid was programmed using either NAL-NL2 or DSL(I/o)v5 prescriptive formula and gain at tinnitus pitch was adjusted till the tinnitus get suppressed.SNR 50 was determined soon after fitted with hearing aid and 30 days of hearing aid use.Further,THI and international outcome inventory for hearing aid(IOI-HA)were determined after 30 days of hearing aid use.A significant higher gain adjustment was needed at tinnitus pitch to reduce tinnitus precept using NAL-NL2 than DSL(I/o)v5 prescriptive formula.Further,SNR 50 was not affected by either tinnitus pitch or revised prescription formulas.However,SNR 50 improved after 30 days of hearing aid use.A 76% of the participants’experienced habituation to perception after 30 days of hearing aid use,10% had slight,10% had mild,and 4% had a moderate degree of tinnitus on THI.On IOA-HA,96%(N=19)of participants have reported satisfactory,and 4%(N=1)reported moderate benefit from hearing aid.Irrespective of prescriptive formula adjusting gain at tinnitus pitch is an efficient method to reduce tinnitus symptoms and improve speech perception.

Subclinical hearing loss associated with aging

Objective:Contribute to clarifying the existence of subclinical hearing deficits associated with aging.Design:In this work,we study and compare the auditory perceptual and electrophysiological performance of normal-hearing young and adult subjects(tonal audiometry,high-frequency tone threshold,a triplet of digits in noise,and click-evoked auditory brainstem response).Study sample:45 normal hearing volunteers were evaluated and divided into two groups according to age.27 subjects were included in the“young group”(mean 22.1 years),and 18 subjects(mean 42.22 years)were included in the“adult group.”Results:In the perceptual tests,the adult group presented significantly worse tonal thresholds in the high frequencies(12 and 16 kHz)and worse performance in the digit triplet tests in noise.In the electrophysiological test using the auditory brainstem response technique,the adult group presented significantly lower I and V wave amplitudes and higher V wave latencies at the supra-threshold level.At the threshold level,we observed a significantly higher latency in wave V in the adult group.In addition,in the partial correlation analysis,controlling for the hearing level,we observed a relationship(negative)between age and speech in noise performance and high-frequency thresholds.No significant association was observed between age and the auditory brainstem response.Conclusion:The results are compatible with subclinical hearing loss associated with aging.

Hearing Disability in Colombia Self-Perception and Associated Factors 2002-2008

Introduction: Hearing disability is a condition that affects normal ear function, as much in adulthood as in the first years of life. According to the 2005 Census, 6.3% of the Colombian population has some type of disability, of which 17.4% have hearing limitations, including those with hearing devices. Elucidating the conditions of this population and identifying the factors related to hearing disability will permit the management of strategies from different sectors to mitigate the consequences associated with this limitation. Objective: To estimate the self-perception of hearing disability in Colombia and to determine the factors associated with this limitation. Materials and Methods: The present study is cross-sectional, based on the analysis of secondary information obtained from the Registry for Localization and Characterization of Persons with Disability (RLCPD) during the 2002-2008 period. Socio-demographic and healthcare variables were analyzed. Results: 13.6% (102,648/750,377) of the population reported hearing limitations even with the use of special hearing devices. 43.52% (44,041) of people over 3 years of age could neither read nor write. 29.39% (30,145) of people who reported this limitation are not affiliated to any sort of health insurance system. Factors found to be associated with hearing limitations were: socioeconomic stratus (OR: 1.33;CI 95% 1.25;1.42), illiteracy (OR: 1.44;CI 95% 1.42;1.46) and lack of affiliation to a health insurance system (OR: 1.03;CI 95% 1.01;1.04). Conclusion: People registered with hearing disability live under vulnerable conditions;among them, most/the majority pertained to a segment of the population with low economic resources and had difficulty obtaining/accessing work, education and healthcare services.

协同感知损失和注意力机制的低剂量CT去噪

由于存在特有的量子噪声,低剂量CT去噪是一项艰巨的任务。当前主流研究使用的深度学习方法存在定性和定量指标不匹配的问题,实验结果的定量指标高,但视觉效果不好。为此,提出一种感知损失和注意力机制的低剂量CT协同去噪网络。该协同机制能够在保证视觉效果的基础上明显改善现有方法定量指标低的问题。模型在网络输入端还引入8方向的边缘检测层,可提取更丰富的纹理与结构信息,进一步提升了网络效果。针对体模数据集和真实临床数据集的实验对比结果表明,该方法相比主流工作,在视觉感受和PSNR以及SSIM指标上,均有更优异表现。

基于改进扩散模型的温度预报

针对传统数值预报模式计算时间长和计算资源消耗大的问题,以及现有深度学习预报方法在温度预报结果上不精确,且预测结果模糊的问题,提出了一个新的温度预报模型。首先,设计了一个时空信息捕捉模块,将该模块捕获的长期依赖信息,作为扩散模型的生成条件,赋予扩散模型预报的能力;其次,设计了一个新的平衡损失函数,同时保护了扩散模型的生成能力和时空信息捕捉模块对时空信息的捕捉能力;最后,基于美国国家环境预报中心的再分析数据进行预报,与现有的深度学习方法相比,所提模型预报结果的质量在均方误差(mean square error,MSE)上降低了17.3%,在均方根误差(root mean square error,RMSE)上降低了9.14%,在峰值信噪比(peak signal to noise ratio,PSNR)上提升了5.1%。改进的扩散模型能有效地捕捉时空依赖的关系,有效地进行时空序列预测,效果优于其他对比方法。

融合动态场景感知和注意力机制的声学回声消除算法

在实时语音频通话系统中,如何去除声学回声得到清晰语音是目前最受关注的难题之一。声学回声消除(Acoustic echo cancellation,AEC)技术旨在消除语音频通话系统中的声学回声,提高通话过程中的语音质量,给予用户良好的通话体验,但是传统回声消除系统存在去回声效果不明显、存在非线性回声残留以及无法实时处理回声等问题。因此,为解决上述存在问题,提出了一种动态场景感知模块(Dynamic scene perception module,DSPM)和全局注意力机制(Global attention mechanism,GAM)相结合的声学回声消除算法。该算法以卷积循环网络(Convolutional recurrent network,CRN)作为基线模型,提取语音信号的序列特征;首先,在其编码器中引入DSPM模块替换原因果卷积,根据场景动态分配卷积内核数量,加强模型的自适应性;其次,在编码器最后两层中分别引入GAM模块,放大空间通道间关系以及统筹全局交互,提升对语音信号特征的提取能力以及消除回声的性能;最后,通过将MSE损失函数和HuberLoss损失函数线性相加生成一种新的损失函数——MSE-HuberLoss,进一步提高模型的鲁棒性。实验结果表明,提出的GAM-DSPM-CRN模型的回声消除性能优秀,且获得较基线模型更加清晰的重构语音信号;在双端通话环境下,提出的GAM-DSPM-CRN模型声学回声消除算法较其他对比算法性能有较大提升;在Microsoft AEC Challenges数据集上,MOS、ERLE和STOI的得分分别达到了4.09、57.43和0.78。

基于虚拟点云的二阶段多模态融合网络

针对点云的稀疏性和无序性对目标检测准确率的影响,本文提出了一种基于虚拟点云的二阶段多模态融合网络VPC-VoxelNet。首先,利用图像检测目标信息构造虚拟点云,增加点云的密集程度,从而提高目标特征的表现;其次,增加点云特征维度以区分真实和虚拟点云,并使用含置信度编码的体素,增强点云的相关性;最后,采用虚拟点云的比例系数设计损失函数,增加图像检测有监督训练,提高二阶段网络训练效率,避免二阶段端到端网络模型存在的模型误差累计问题。该目标检测网络VPC-VoxelNet在KITTI数据集上进行了测试,检测精度优于经典三维点云检测网络和某些多传感器信息融合网络,车辆检测精度达到了86.9%。

基于感知自编码器的军品电路表面缺陷检测方法

面向军品电路的表面缺陷检测任务,由于军品电路存在多品种、小批量,表面复杂的特点,现有方法对图像重建效果较差,本研究提出一种基于感知自编码器(Perceptual AutoEncoder,PAE)的方法,将感知损失与自注意力模块引入无监督方法,增加方法的可迁移性与图像重建效果。与传统检测方法相比,基于感知自编码器的方法无需面临传统模板法的对齐、光照平衡、色彩平衡等问题,极大地提升了针对不同产品的可迁移性,可有效解决军品电路多品种、小批量检测面临的困难。具体方法为:使用特征金字塔与卷积方法提取不同尺度的特征向量并聚类,聚类后使用自注意力模块自动加权并增强需要关注的特征,而后重建图像,将该图像作为模板与输入进行差分比较。针对感知自编码器,在自制的数据集上进行了评估,评估结果表明,引入感知损失后的自编码器能够更准确地进行缺陷检测。

不同部位注射NGF治疗突发性耳聋对患者听力及症状的影响

【目的】探讨不同部位注射鼠神经生长因子(NGF)治疗突发性耳聋对患者听力及症状的影响。【方法】选取2019年10月至2022年10月韩城市人民医院收治的70例突发性耳聋患者,按照随机数字表法分为观察组和对照组,每组35例。对照组采用臀部注射NGF治疗,观察组采用耳后乳突骨膜下注射NGF治疗。比较两组患者临床疗效、不良反应发生率、临床症状(包括耳鸣、耳闷、眩晕)消失时间及治疗前后听阈值(包括骨传导、空气传导听阈值)。【结果】观察组临床疗效总有效率为94.29%,显著高于对照组的74.29%,差异有统计学意义(P<0.05)。观察组耳鸣、耳闷、眩晕消失时间均短于对照组,差异有统计学意义(P<0.05)。治疗后,两组患者骨传导、空气传导听阈值均低于同组治疗前,差异有统计学意义(P<0.05);且观察组治疗后骨传导、空气传导听阈值均低于对照组治疗后,差异有统计学意义(P<0.05)。两组患者不良反应总发生率比较,差异无统计学意义(P>0.05)。【结论】与臀部注射NGF相比,耳后乳突骨膜下注射NGF治疗突发性耳聋临床疗效更好,患者临床症状改善更明显,促进患者听力恢复,且安全性较高。

Auditory perceptual efficacy of nonlinear frequency compression used in hearing aids:A review

Many patients with sensorineural hearing loss have a precipitous high-frequency loss with relatively good thresholds in the low frequencies.This present paper briefly introduces and compares the basic principles of four types of frequency lowering algorithms with emphasis on nonlinear frequency compression(NLFC). A review of the effects of the NLFC algorithm on speech and music perception and sound quality appraisal is then provided. For vowel perception, it seems that the benefits provided by NLFC are limited, which are probably related to the parameter settings of the compression. For consonant perception, several studies have shown that NLFC provides improved perception of highfrequency consonants such as/s/and/z/. However, a few other studies have demonstrated negative results in consonant perception. In terms of sentence recognition, persistent use of NLFC might provide improved performance. Compared to the conventional processing, NLFC does not alter the speech sound quality appraisal and music perception as long as the compression setting is not too aggressive. In the subsequent section,the relevant factors with regard to NLFC settings, time-course of acclimatization, listener characteristics, and perceptual tasks are discussed.Although the literature shows mixed results on the perceptual efficacy of NLFC, this technique improved certain aspects of speech understanding in certain hearing-impaired listeners. Little research is available on speech perception outcomes in languages other than English. More clinical data are needed to verify the perceptual efficacy of NLFC in patients with precipitous high-frequency hearing loss. Such knowledge will help guide clinical rehabilitation of those patients.

基于标签嵌入方法的纺织品瑕疵识别网络

卷积神经网络(Convolutional Neural Network,CNN)可用于工业生产环境下的纺织品疵点的鉴别与分类。针对实际场景下的纺织品瑕疵存在瑕疵类型视觉区分度小和实际数据样本采集时的瑕疵类别不平衡问题,本文提出了基于标签嵌入方法的纺织品瑕疵识别网络(Textile Defect Recognition Network Based on Label Embedding,TDRNet)。首先,算法调整了基础骨干网络的结构,从而提高模型的分类精度;接着算法还设计了标签嵌入模块(Label Embedded Module,LEM),并使用该模块来生成模型的类别权重偏移;然后,本文提出了分布感知损失函数(Distribution Perception Loss,DP Loss)调整算法的类别分布,从而减小同类瑕疵特征的类内距并增大异类瑕疵特征的类间距;最后,本文引入了Seesaw Loss损失函数,通过抑制少数类别的负样本梯度并提高对误分类时的样本损失来动态平衡模型训练过程中在不同样本下的更新梯度,以缓解少数类别的误分类率。在自制的“广东智能制造”布匹瑕疵分类数据集中,本文提出的框架在粗粒度分类和细粒度分类两个任务上的top1错误率可达16.35%和17.12%,而top5错误率在细粒度分类任务上低至5.20%。与其他分类模型相比,TDRNet在对比实验中取得了最优的结果。此外,TDRNet与近5年经典的细粒度分类模型进行了比较,并取得了SOTA结果,这充分表明了TDRNet的先进性。

基于身份感知模型的素描人脸识别方法

针对素描图像和光学图像之间存在较大的模态差异这一问题,提出了一种基于身份感知模型的素描人脸识别方法,实现跨模态图像生成和素描人脸识别。该方法应用新的感知损失来监督图像生成网络,生成更好的跨模态图像,减少模态差异带来的识别精度损失,并通过三元组损失来正则化类内和类间距离,增强识别模型的性能,用联合训练策略提升素描人脸识别能力。在UoM-SGFSv2、e-PRIP等素描人脸数据集上的实验结果表明,该方法识别效果优于其他对比算法。

基于自适应特征感知的轻量化人体姿态估计

针对现有人体姿态估计网络在追求高精度检测时,网络结构设计复杂、模型参数量较大、检测效率较低的问题,本文提出了一种基于自适应特征感知的轻量级人体姿态估计算法。首先利用轻量化Ghost模块重构人体姿态估计的特征提取网络,减少网络参数量;其次设计了一种轻量级自适应特征感知的注意力机制,在降低网络模型复杂度的同时加强通道间信息有效交流,有效改善关键点定位效果;最后采用Huber Loss损失函数优化模型训练,实现异常点的更优预测,增强模型鲁棒性。在COCO数据集上进行验证,实验结果表明,与基准RMPE算法相比,改进后模型的检测精度提升了约0.5%,参数量减少了56.0%,网络运算量降低了32.6%,模型体积压缩了约57.0%,模型检测速率提升约2.1倍。本文改进后的人体姿态估计模型在压缩模型体积的同时提高了检测效率,增强了模型鲁棒性。

基于高斯分布建模的序列推荐算法

目前大多数序列推荐系统(SRS)都假设需要预测的下一项与用户的上一项输入有关,然而真实场景中,用户可能会在浏览过程中误点击与自身兴趣偏好不一致的项目(不可靠实例)。针对此问题,提出了一种基于高斯分布建模的序列推荐算法。该算法首先通过一个含有多头自注意力的不确定性感知图集合网络(uncertainty-aware graph ensemble network,UAN),通过降低输入项的不确定性来提取输入的序列模式;其次将提取的输入序列模式建模为一个高斯分布,得到序列信息中的动态用户偏好以及偏好的不确定性;再将传统的推荐目标函数拓展为一个采样损失函数和一个不确定性正则化器,赋予每个训练实例适当的不确定性;最后将高损失且低不确定性的不可靠实例去除,增强序列推荐的准确性。该算法在三个公开的数据集Book-Crossing、MovieLens-1M和Steam上进行实验测试,结果表明,该算法相对于效果较好的基线取得了5.3%左右的提升,得到了更优的序列推荐结果,并能通过有效降低输入序列信息的不确定性,从而提升推荐准确率。

改进YOLOv5s的复杂交通场景路侧目标检测算法

针对传统路侧目标检测模型存在的对于行人、非机动车、受遮挡车辆等小目标检测精度低以及模型体积过大的问题,提出了一种基于改进YOLOv5s的路侧目标检测模型。使用EIoU Loss替换原始的CIoU Loss作为目标边界框的回归损失函数,在加快预测框回归损失函数收敛速度的同时提升了预测框的回归预测精度;使用轻量级的通用上采样算子CARAFE替换原始的最近邻插值上采样模块,减少了上采样过程中特征信息的损失;在原始的三尺度检测层的基础上新添加一层检测尺度更小的小目标检测分支,并提出了一种高效的解耦预测头对不同尺度的检测层进行解耦,进一步提升了模型对于小目标的检测能力;对改进后的模型进行通道剪枝,剪除对于检测效果影响不大的冗余通道,降低模型体积,使得模型更加适用于资源受限条件下的路侧目标检测任务。在路侧目标检测数据集DAIR-V2X-Ⅰ上的实验结果表明,相较于原始YOLOv5s算法,改进后的算法在模型体积减小5.7 MB的基础上,mAP50、mAP50:95分别提高了2.5个百分点和3.8个百分点,达到了90.3%、67.7%,检测速度也达到了89 FPS。与其他主流的目标检测算法在检测精度、模型体积以及检测速度上相比有一定的优势,改进后的算法适用于复杂交通场景下的路侧目标检测任务。

基于字形感知和注意力归一化的字体迁移

字体迁移是一项十分具有挑战性的任务,其目的是将目标字体通过某种映射方式迁移到源字体,以实现字体的变换。现有的方法在字体迁移方面的鲁棒性有限,突出表现为对生成字体结构完整性的保持较差,尤其是当两种不同种类的字体差别较大时。针对这些问题,提出了一种端到端的字体迁移网络框架模型。该模型引入了注意力归一化以更好地提取字形图像的高级语义特征,从而提高生成图像的质量。此外,使用自适应实例归一化进行字体特征和内容特征融合,以实现字体的转换。在保持字形结构完整性方面,设计了感知损失和上下文损失来约束字形结构的生成。为了稳定GAN网络的训练,在对抗损失函数的设计中加入了正则化项。为了验证该模型的有效性,实验采用FET-GAN中公开的数据集进行了多组训练和测试,并与FET-GAN,CycleGAN和StarGANv2进行了对比。实验结果表明,该模型能够在给定的多个字体域之间实现相互的字体迁移,并且其迁移的效果和模型泛化能力与其他工作相比均具有一定的优势。

面向分类和深度损失的立体图像重定向质量评价

现有的立体重定向图像客观评价算法往往采用统一的方法对重定向图像的失真予以衡量.然而,对于不同类型的立体重定向图像,人眼关注的失真类型与模式不尽相同.此外,现有算法未能精确地衡量重定向图像中深度感变化.针对上述问题,本文提出了一种基于分类和深度感损失的立体图像重定向质量评价算法.该算法根据重定向图像的显著度和深度特征进行图像分类,对不同类型的图像采用不同的失真质量评价策略.此外,为了精确衡量重定向图像的深度感变化,本文还提出了两种用于衡量深度感损失的指标,分别利用重定向过程中深度信息的丢失以及绝对深度的相似度来衡量深度感的损失.本文利用NBU-SIRQA和SIRD数据库进行实验,结果表明,本文算法的整体评价性能优于现有所有算法.

Temporal Convolutional Network for Speech Bandwidth Extension

In the field of speech bandwidth exten-sion,it is difficult to achieve high speech quality based on the shallow statistical model method.Although the application of deep learning has greatly improved the extended speech quality,the high model complex-ity makes it infeasible to run on the client.In order to tackle these issues,this paper proposes an end-to-end speech bandwidth extension method based on a temporal convolutional neural network,which greatly reduces the complexity of the model.In addition,a new time-frequency loss function is designed to en-able narrowband speech to acquire a more accurate wideband mapping in the time domain and the fre-quency domain.The experimental results show that the reconstructed wideband speech generated by the proposed method is superior to the traditional heuris-tic rule based approaches and the conventional neu-ral network methods for both subjective and objective evaluation.

视觉感知下的无人机拍摄图像最邻近帧质量增强

针对无人机拍摄图像邻近帧清晰度不达标的问题,为提升图像帧节点的质量水平,提出视觉感知下的无人机拍摄图像最邻近帧质量增强方法。以视觉感知模型为基础,确定定量感知参数取值范围,再通过求解回归判定向量的方式,实现对无人机拍摄图像的质量评价。根据邻近帧结构相似性损失表达式,计算像素亮度残差,联合相关参数指标,定义定量增强函数条件,完成视觉感知下无人机拍摄图像最邻近帧质量增强。实验结果表明,所提算法作用下的无人机拍摄图像分辨率最大值为180 ppi,满足提升无人机拍摄图像帧节点质量水平、解决图像清晰度不达标问题的实际应用需求。

基于FGCM的通信网络运行态势感知模型研究

为有效降低不同通信网络链路中的运行带宽与数据丢包率,保障网络体系的安全运行态势,文中提出一种基于FGCM的通信网络运行态势感知模型。依据Flink通信协议连接需求定义因子集表达式;再联合权重向量,根据FGCM算法评价原则对通信网络运行数据进行评价,提取通信网络的态势特征;最后,根据网络运行指数的取值范围,估算主机元件对通信数据参量的感知能力,实现通信网络运行态势感知模型。实验结果表明,所提方法可以有效降低网络运行带宽和数据丢包率,保障网络体系的安全运行态势。