Nano-volt magnetic resonance sounding(MRS) signals are sufficiently weak so that during the actual measurement, they are affected by environmental electromagnetic noise, leading to inaccuracy of the extracted characte...Nano-volt magnetic resonance sounding(MRS) signals are sufficiently weak so that during the actual measurement, they are affected by environmental electromagnetic noise, leading to inaccuracy of the extracted characteristic parameters and hindering effective inverse interpretation. Considering the complexity and non-homogeneous spatial distribution of environmental noise and based on the theory of adaptive noise cancellation, a model system for noise cancellation using multi-reference coils was constructed to receive MRS signals. The feasibility of this system with theoretical calculation and experiments was analyzed and a modified sigmoid variable step size least mean square(SVSLMS) algorithm for noise cancellation was presented. The simulation results show that, the multi-reference coil method performs better than the single one on both signal-to-noise ratio(SNR) improvement and signal waveform optimization after filtering, under the condition of different noise correlations in the reference coils and primary detecting coils and different SNRs. In particular, when the noise correlation is poor and the SNR<0, the SNR can be improved by more than 8 dB after filtering with multi-reference coils. And the average fitting errors for initial amplitude and relaxation time are within 5%. Compared with the normalized least mean square(NLMS) algorithm and multichannel Wiener filter and processing field test data, the effectiveness of the proposed method is verified.展开更多
The basic idea of multi-class classification is a disassembly method,which is to decompose a multi-class classification task into several binary classification tasks.In order to improve the accuracy of multi-class cla...The basic idea of multi-class classification is a disassembly method,which is to decompose a multi-class classification task into several binary classification tasks.In order to improve the accuracy of multi-class classification in the case of insufficient samples,this paper proposes a multi-class classification method combining K-means and multi-task relationship learning(MTRL).The method first uses the split method of One vs.Rest to disassemble the multi-class classification task into binary classification tasks.K-means is used to down sample the dataset of each task,which can prevent over-fitting of the model while reducing training costs.Finally,the sampled dataset is applied to the MTRL,and multiple binary classifiers are trained together.With the help of MTRL,this method can utilize the inter-task association to train the model,and achieve the purpose of improving the classification accuracy of each binary classifier.The effectiveness of the proposed approach is demonstrated by experimental results on the Iris dataset,Wine dataset,Multiple Features dataset,Wireless Indoor Localization dataset and Avila dataset.展开更多
目的利用360°全方向24和36声源测试设备,初步探讨健听中青年和健听老年前期-老年人水平声源定位特点。方法选取2021年4月至2021年9月中国人民解放军总医院耳鼻喉科收治的43例健听成年受试者为研究对象,其中男性22例,女性21例;根据...目的利用360°全方向24和36声源测试设备,初步探讨健听中青年和健听老年前期-老年人水平声源定位特点。方法选取2021年4月至2021年9月中国人民解放军总医院耳鼻喉科收治的43例健听成年受试者为研究对象,其中男性22例,女性21例;根据年龄分为中青年组(21~49岁)20例和老年前期-老年组(50~72岁)23例。两组分别给予纯音听阈测试、全方向24声源(间隔15°)和36声源(间隔10°)水平声源定位(sound localization,SL)能力评估。给声强度60 dB HL,给声刺激为1 kHz啭音,通过计算均方根误差(root mean square,RMS)、平均绝对误差(mean absolutely error,MAE)等评估受试者的声源定位能力。结果24声源老年前期-老年组MAE、RMS均值高于中青年组的MAE、RMS均值,差异有统计学意义(P<0.05);36声源老年前期-老年组MAE、RMS高于中青年组的MAE、RMS,差异无统计学意义(P>0.05)。24声源和36声源前场MAE和RMS均高于后场的MAE和RMS,前后场的MAE和RMS比较,差异有统计学意义(P<0.01);左右场的MAE、RMS比较,差异无统计学意义(P>0.05)。24声源前后混淆比例为7.73%,36声源前后混淆比例为15.42%;24声源和36声源均为正前方的声源定位准确度最差;老年前期-老年组前后混淆的比例高于中青年组,差异无统计学意义(P>0.05)。结论健听老年前期-老年人全方向24声源和36声源水平定位能力,相比健听中青年组有所下降。左右场的定位准确度高,前后场的定位准确度低,正前方定位准确度最低。全方向水平声源定位能力的测试结果与扬声器数量有关,且反应趋势具有一致性。展开更多
波达方向(Direction of Arrival,DOA)估计技术是语音增强和声学探测中的重要工具,对于语音机器人、视频会议、助听器和声呐等应用至关重要。最近出现的DOA估计新方法,例如图信号处理(Graph Signal Processing,GSP)方法,展现出优异的角...波达方向(Direction of Arrival,DOA)估计技术是语音增强和声学探测中的重要工具,对于语音机器人、视频会议、助听器和声呐等应用至关重要。最近出现的DOA估计新方法,例如图信号处理(Graph Signal Processing,GSP)方法,展现出优异的角度估计能力,有望提供更佳的声源DOA估计解决方案。然而,由于在多声源情况下GSP算法由邻接矩阵无法直接得到接收信号特征向量的正交补矩阵,导致多声源下GSP算法失效。为解决此问题,本文基于多源宽带语音信号的频域单源区域检测实现多声源分离,进而利用GSP和聚类算法实现宽带多声源的定位。具体而言,本文首先将GSP方法扩展到频域。其次,利用短时傅里叶变换将信号分为若干时频区域,筛选出单源主导的时频区域后,对其进行频域GSP单源定位。最后,对所有定位结果进行聚类,再通过加权平均获得最终的角度估计。我们利用LibriSpeech语音语料库构建声源信号进行多声源定位仿真,仿真结果证明,本文方法优于其他算法,较高信噪比下可将误差控制在3°以内。此外,我们使用圆形六阵元麦克风阵列,对实际录制的若干组录音数据应用所提算法进行定位测量,结果展示所提算法的定位误差更小,并在声源较为靠近时也能做到较好的分辨。展开更多
基金Projects(41204079,41504086)supported by the National Natural Science Foundation of ChinaProject(20160101281JC)supported by the Natural Science Foundation of Jilin Province,ChinaProjects(2016M590258,2015T80301)supported by the Postdoctoral Science Foundation of China
文摘Nano-volt magnetic resonance sounding(MRS) signals are sufficiently weak so that during the actual measurement, they are affected by environmental electromagnetic noise, leading to inaccuracy of the extracted characteristic parameters and hindering effective inverse interpretation. Considering the complexity and non-homogeneous spatial distribution of environmental noise and based on the theory of adaptive noise cancellation, a model system for noise cancellation using multi-reference coils was constructed to receive MRS signals. The feasibility of this system with theoretical calculation and experiments was analyzed and a modified sigmoid variable step size least mean square(SVSLMS) algorithm for noise cancellation was presented. The simulation results show that, the multi-reference coil method performs better than the single one on both signal-to-noise ratio(SNR) improvement and signal waveform optimization after filtering, under the condition of different noise correlations in the reference coils and primary detecting coils and different SNRs. In particular, when the noise correlation is poor and the SNR<0, the SNR can be improved by more than 8 dB after filtering with multi-reference coils. And the average fitting errors for initial amplitude and relaxation time are within 5%. Compared with the normalized least mean square(NLMS) algorithm and multichannel Wiener filter and processing field test data, the effectiveness of the proposed method is verified.
基金supported by the National Natural Science Foundation of China(61703131 61703129+1 种基金 61701148 61703128)
文摘The basic idea of multi-class classification is a disassembly method,which is to decompose a multi-class classification task into several binary classification tasks.In order to improve the accuracy of multi-class classification in the case of insufficient samples,this paper proposes a multi-class classification method combining K-means and multi-task relationship learning(MTRL).The method first uses the split method of One vs.Rest to disassemble the multi-class classification task into binary classification tasks.K-means is used to down sample the dataset of each task,which can prevent over-fitting of the model while reducing training costs.Finally,the sampled dataset is applied to the MTRL,and multiple binary classifiers are trained together.With the help of MTRL,this method can utilize the inter-task association to train the model,and achieve the purpose of improving the classification accuracy of each binary classifier.The effectiveness of the proposed approach is demonstrated by experimental results on the Iris dataset,Wine dataset,Multiple Features dataset,Wireless Indoor Localization dataset and Avila dataset.
文摘目的利用360°全方向24和36声源测试设备,初步探讨健听中青年和健听老年前期-老年人水平声源定位特点。方法选取2021年4月至2021年9月中国人民解放军总医院耳鼻喉科收治的43例健听成年受试者为研究对象,其中男性22例,女性21例;根据年龄分为中青年组(21~49岁)20例和老年前期-老年组(50~72岁)23例。两组分别给予纯音听阈测试、全方向24声源(间隔15°)和36声源(间隔10°)水平声源定位(sound localization,SL)能力评估。给声强度60 dB HL,给声刺激为1 kHz啭音,通过计算均方根误差(root mean square,RMS)、平均绝对误差(mean absolutely error,MAE)等评估受试者的声源定位能力。结果24声源老年前期-老年组MAE、RMS均值高于中青年组的MAE、RMS均值,差异有统计学意义(P<0.05);36声源老年前期-老年组MAE、RMS高于中青年组的MAE、RMS,差异无统计学意义(P>0.05)。24声源和36声源前场MAE和RMS均高于后场的MAE和RMS,前后场的MAE和RMS比较,差异有统计学意义(P<0.01);左右场的MAE、RMS比较,差异无统计学意义(P>0.05)。24声源前后混淆比例为7.73%,36声源前后混淆比例为15.42%;24声源和36声源均为正前方的声源定位准确度最差;老年前期-老年组前后混淆的比例高于中青年组,差异无统计学意义(P>0.05)。结论健听老年前期-老年人全方向24声源和36声源水平定位能力,相比健听中青年组有所下降。左右场的定位准确度高,前后场的定位准确度低,正前方定位准确度最低。全方向水平声源定位能力的测试结果与扬声器数量有关,且反应趋势具有一致性。
文摘波达方向(Direction of Arrival,DOA)估计技术是语音增强和声学探测中的重要工具,对于语音机器人、视频会议、助听器和声呐等应用至关重要。最近出现的DOA估计新方法,例如图信号处理(Graph Signal Processing,GSP)方法,展现出优异的角度估计能力,有望提供更佳的声源DOA估计解决方案。然而,由于在多声源情况下GSP算法由邻接矩阵无法直接得到接收信号特征向量的正交补矩阵,导致多声源下GSP算法失效。为解决此问题,本文基于多源宽带语音信号的频域单源区域检测实现多声源分离,进而利用GSP和聚类算法实现宽带多声源的定位。具体而言,本文首先将GSP方法扩展到频域。其次,利用短时傅里叶变换将信号分为若干时频区域,筛选出单源主导的时频区域后,对其进行频域GSP单源定位。最后,对所有定位结果进行聚类,再通过加权平均获得最终的角度估计。我们利用LibriSpeech语音语料库构建声源信号进行多声源定位仿真,仿真结果证明,本文方法优于其他算法,较高信噪比下可将误差控制在3°以内。此外,我们使用圆形六阵元麦克风阵列,对实际录制的若干组录音数据应用所提算法进行定位测量,结果展示所提算法的定位误差更小,并在声源较为靠近时也能做到较好的分辨。
