一种用于立体声声学回波消除的新型鲁棒梯度法格梯形自适应滤波算法

提出了一种使用编码器−时序建模结构的时延估计方法来估计声学回声抵消中传声器信号相对远端信号的时延。该方法以短时傅里叶变换域的远端信号和传声器信号作为输入特征, 通过复数卷积神经网络构成的编码器提取带有相位信息的高维特征, 利用循环神经网络学习两输入信号之间的时延关系, 构建了从信号到时延的映射。仿真实验结果表明, 相比WebRTC-DE和GCC-PHAT, 所提方法的优势有: (1)模型的参数量和计算量不受时延长度影响; (2)有效缩短了时延估计的收敛时间和跟踪时间; (3)在长混响和双端对讲的情况下具有更小、更稳定的估计误差和标准差。将使用编码器−时序建模结构的时延估计方法与自适应回声抵消级联的实验验证了新方法的有效性。

     

A two-stage convolutional recurrent network-based stereophonic acoustic echo cancellation approach with complex spectrum

A two-stage convolutional recurrent network(CRN) with complex spectral input features is proposed to address the stereophonic acoustic echo cancellation(SAEC) problem.The proposed algorithm avoids the decorrelation of far-end signals, which solves the non-unique solution problem of the adaptive filter-based SAEC and ensures the stereo sound quality and spatial perception. It deals with SAEC problem in two stages. In the first stage, a CRN model is used to estimate the echo signal based on the mi...     

门控卷积神经网络多通路声学回声消除算法*

多通路声重放系统能够增强听者的现实感与空间感,但在免提通信条件下,其不可避免会受到噪声和回声干扰,严重影响通信质量。针对上述问题,本文提出了一种基于门控卷积循环神经网络的多通路声学回声消除和噪声抑制方法。该方法以传声器接收信号和重放声道的压缩复数谱为网络输入,以近端语音的压缩复数谱为网络的输出目标,直接从传声器拾取信号中恢复近端纯净语音,无需对声重放信号进行去相关处理,解决了传统自适应滤波方法中存在的非唯一解问题,同时保证了多通路声重放质量。仿真和真实声学环境实验均表明本文所提出的方法可显著消除多通路声重放系统的噪声和回声,在语音质量和回声返回衰减增益方面均优于传统算法。     

双通道回声抵消系统中改进算法的定点化实现

本文研究了双通道回声抵消系统的三个核心模块:频域多延时自适应滤波器算法(MDF),双端说话检测算法(DTD),以及残留回声抑制算法(RES)。针对频域算法的特点提出了改进的双端说话检测算法和基于维纳滤波与谱减法的残留回声抑制算法,然后以上述三种算法为核心模块实现双通道回声抵消系统,并对系统定点化以便在定点DSP处理器上实时实现,分析并解决了定点化的精度问题所带来的影响。     

声学回波统计的鱼群密度评估方法*

与传统的回波积分法相比，回波统计鱼群数量密度评估方法具有无需回波信号绝对量和鱼群中个体鱼的平均目标强度先验知识的优点。对该方法在鱼群数量密度评估上的应用问题进行了实验研究，并针对其中鱼群密度分布的不均匀性及统计样本量偏少导致评估结果偏差的问题，提出了一种基于能量阈值的鱼群回波数据预处理方法和统计样本的改进抽样方法，有效提高了该方法鱼群密度评估的精确度。与传统的回波积分法评估结果的对比分析表明，该评估方法适用于低密度鱼群的评估。     

Event-based state estimation for a class of complex networks with time-varying delays: A comparison principle approach

The paper is concerned with the state estimation problem for a class of time-delayed complex networks with event-triggering communication protocol. A novel event generator function, which is dependent not only on the measurement output but also on a predefined positive constant, is proposed with hope to reduce the communication burden. A new concept of exponentially ultimate boundedness is provided to quantify the estimation performance. By means of the comparison principle, some sufficient conditions are obtained to guarantee that the estimation error is exponentially ultimately bounded, and then the estimator gains are obtained in terms of the solution of certain matrix inequalities. Furthermore, a rigorous proof is proposed to show that the designed triggering condition is free of the Zeno behavior. Finally, a numerical example is given to illustrate the effectiveness of the proposed event-based estimator.