运动平台混响背景下基于低秩稀疏分解的目标回波增强

在水下航行器等运动平台上,主动声呐的近距离滤波结果受混响干扰影响严重,大量的混响回波亮点会掩蔽目标回波的可见性,导致后续检测判决的虚警率增大。以阵列处理的方位历程图作为基本输入,该文利用某些场景下混响干扰相邻周期间潜在的相干结构,假设混响满足低秩性;由于平台间的相对运动,假设感兴趣的目标回波在逐周期间是不相关且稀疏的。之后,将方位历程图表示为低秩的混响、稀疏的运动目标回波和噪声成分,在此基础上提出以加速近端梯度法(APG)和快速数据投影法(FDPM)分别实现离线和在线的低秩稀疏分解,从而实现混响抑制和目标回波增强。试验结果验证了假设模型的有效性,并且两种分解算法均能有效地增强目标回波。     

运动平台混响背景下基于低秩稀疏分解的目标回波增强

在水下航行器等运动平台上,主动声呐的近距离滤波结果受混响干扰影响严重,大量的混响回波亮点会掩蔽目标回波的可见性,导致后续检测判决的虚警率增大.以阵列处理的方位历程图作为基本输入,该文利用某些场景下混响干扰相邻周期间潜在的相干结构,假设混响满足低秩性;由于平台间的相对运动,假设感兴趣的目标回波在逐周期间是不相关且稀疏的.之后,将方位历程图表示为低秩的混响、稀疏的运动目标回波和噪声成分,在此基础上提出以加速近端梯度法(APG)和快速数据投影法(FDPM)分别实现离线和在线的低秩稀疏分解,从而实现混响抑制和目标回波增强.试验结果验证了假设模型的有效性,并且两种分解算法均能有效地增强目标回波.     

改进的多级线性预测晚期混响抑制算法

在一个封闭的空间内,距离声源较远时接收到的语音信号通常会被混响所污染,其中晚期混响会在很大程度上降低语音可懂度。一般的去噪方法只能去除常见的加性噪声如白噪声,并不能去除房间冲激响应与干净语音卷积而成的混响,因此需要专门的去混响算法来去除晚期混响带来的影响。本文提出了一种新算法,在多级线性预测单通道去混响算法的基础上,修正了其预白化过程,改进后的算法可以提升语音前两个共振峰。实验结果证明,新算法在去除大部分混响的同时能够保留更多的有用语音的低频成分,因而提高了语音可懂度。      

基于小波变换的声学回波消除系统

声学回波消除技术一直是语音通信领域的研究热点。在声学回波消除系统中,通过估计回波路径中的固定时延区域来提高自适应滤波算法的收敛速度。提出了一种基于小波变换的固定时延估计算法以及基于小波变换的声学回波消除系统,解决传统时延估计算法在声学回波消除系统中估计误差大、抗干扰能力弱的问题。测试结果表明,算法稳健性、有效性等指标明显优于传统时延估计算法,基于小波变换的声学回波消除系统具有良好的消回波性能。     

主动声纳中混响干扰的一种非线性抑制方法

本文介绍了一种基于混沌理论的非线性动力学信号分离方法——变尺度概率净化法,并将之优化应用于主动声纳目标检测中混响干扰的抑制。该方法以近似条件下不含目标回波的纯混响数据作为参考信号,使目标回波分离后的混响背景数据与参考混响数据在相空间上具有一致的概率分布,使目标回波与混响背景适度分离,从而实现对混响干扰的抑制。在信混比较高的情况下,该方法具有较好的混响抑制效果。另外,论文对算法作了一些改进,降低了算法的运算量和运行所需内存,而分离性能并没有下降。     

国内要闻

声学所与长虹集团联合研发成功中国首款复合型智能语音芯片近日,中国首款复合型智能语音芯片由中科院语言声学与内容理解重点实验室和长虹集团联合研发成功。复合型语音智能芯片突出优势是在语音识别技术的基础上,融合了噪声和混响抑制、回波消除、波束成形等多方面的语音增强技术,将改变现有智能语音家电采用手持遥控器控制的方式,实现远距离语音控制。基于长虹集团黑白融合的全产品线,这款智能语音芯片有望成为中国智能语音市场标志性芯片产品。     

融合注意力机制的CS-BiLSTM深度回声消除算法

在全双工通信系统中，声学回声会降低用户的体验，针对在双向通话场景下自适应滤波算法消除声学回声效果不理想以及非线性声学回声难以消除的问题，提出一种注意力机制与BiLSTM网络相结合的CS-BiLSTM深度声学回声消除算法。首先通过构建BiLSTM网络提取语音的时序特征，之后引入通道和空间注意力机制提取回声信号的空间特征信息，并融合均方根误差与平均绝对误差提出一种新的损失函数，提高模型的鲁棒性。改进后的CS-BiLSTM网络模型能够获得清晰的语音信号，具有更好的回声消除性能。仿真结果表明，在非线性回声和双向通话环境下，与其他几种参考算法相比，所提出的CS-BiLSTM算法在感知语音质量评价方面明显优于其他算法，更有效地实现了回声消除，此外，该算法结构简单且模型参数量更少。     

基于最小相位分解的语音去混响

为解决语音通信中的混响干扰问题,提出了一种基于最小相位分解的,可用于单通道语音的去混响方法。根据信号最小相位分解的原理,将接收到的含噪带混响的语音信号分解成最小相位部分和全通部分,对其中的最小相位部分进行复倒谱域的滤波处理,再与全通部分进行合成以实现混响的去除。通过仿真实验来实现这种方法,并与另一种改进过的复倒谱域滤波的去混响方法进行比较,实验结果表明这种方法相对较好。     

基于深度学习特征融合和联合约束的单通道语音分离方法

为了提高单通道语音分离性能,该文提出基于深度学习特征融合和联合约束的单通道语音分离方法。传统基于深度学习的分离算法的损失函数只考虑了预测值和真实值的误差,这使得分离后的语音与纯净语音之间误差较大。该文提出一种新的联合约束损失函数,该损失函数不仅约束了理想比值掩蔽的预测值和真实值的误差,还惩罚了相应幅度谱的误差。另外,为了充分利用多种特征的互补性,提出一种含特征融合层的卷积神经网络(CNN)结构。利用该CNN提取多通道输入特征的深度特征,并在融合层中将深度特征与声学特征融合用来训练分离模型。由于融合构成的特征含有丰富的语音信息,具有强的语音信号表征能力,使得分离模型预测的掩蔽更加准确。实验结果表明,从信号失真比(SDR) 、主观语音质量评估( PESQ)和短时客观可懂度(STOI)3个方面评价,相比其他优秀的基于深度学习的语音分离方法,该方法能够更有效地分离目标语音。     

弹性物体在混响背景下的散射信号检测

混响是主动声呐的主要背景干扰,是由发射信号引起的,其谱结构与发射信号也具有一定的相似性,是一种非平稳的有色干扰噪声.特别是在浅水情况下,强大的混响干扰尤其严重.而在实验数据中有时并没有获得发射声信号的取样.本文采取一种尝试,截取一段高信混比回波作为发射信号来考虑,同时截取纯混响叠加到回波信号上,以模拟低信混比的情况,在混响自回归(AR)模型的基础上研究混响的预白化处理,并将预白化处理后的信号用于匹配滤波器的信号检测.     

基于BLSTM神经网络的回声和噪声抑制算法

考虑到非线性回声和非平稳噪声对智能设备回声消除算法的影响，论文提出一种基于双向长短时记忆(Bidirectional Long Short-Term Memory，BLSTM)神经网络的回声和噪声抑制算法。该算法首先采用多目标预处理模型，同步估计出回声和噪声信号的幅度谱；然后将其作为回声和噪声抑制模型的输入特征，进而估计出目标语音信号的理想比例掩模；最后通过联合训练两个模型得到最优回声和噪声抑制模型。实验结果表明，在非线性回声和非平稳噪声的环境下，该算法均取得了较好的回声和噪声抑制效果，语音失真较小。      

Gammatone滤波器修正的多级线性预测去混响

在相对封闭的空间内,使用一些通讯设备时,距离声源较远的麦克风接收到的语音信号通常会被混响和噪声所污染,这些干扰信号会在很大程度上降低语音可懂度。带噪的混响语音经过去混响和消噪算法处理后,常常会残留一些音乐噪声,使得人耳听起来很不舒服。本文根据人耳对不同频带的听觉特征,引入Gammatone听觉滤波器组,提出了Gammatone滤波器修正的多级线性预测去混响算法。实验结果证明,新算法有效的解决了去混响及消噪后的残留音乐噪声问题,提高了语音的清晰度和舒适度。      

Robust Voice Activity Detection Based on Concept of Modulation Transfer Function in Noisy Reverberant Environments

Voice activity detection (VAD) is used to detect speech and non-speech periods from observed speech signals. It is an important front-end technique for many speech technology applications. Many VAD methods have been proposed. However most of them have been applied under clean or noisy conditions. Only a few methods have been proposed for reverberant conditions, particularly under noisy reverberant conditions. We therefore need to understand the ill effects of noise and reverberation on speech to design an accurate and robust method of VAD under noisy reverberant conditions. The ill effects of noise and reverberation for speech can be regarded as the modulation transfer function (MTF) under noisy and reverberant conditions. Therefore, our study is based on the MTF concept to reduce the ill effects of noise and reverberation on speech, and propose a robust VAD method that we obtained in this study. Noise reduction and dereverberation were first applied to the temporal power envelope of the speech signal to restore the temporal power envelope with this method. Then, power thresholding as a VAD decision was designed based on the restored temporal power envelope. A method of estimating the signal to noise ratio (SNR) was proposed to accurately estimate the SNR in the noise reduction stage. Experiments under both artificial and realistic noisy reverberant conditions were carried out to evaluate the performance of the proposed method of VAD and it was compared with conventional VAD methods. The results revealed that the proposed method significantly outperformed the conventional methods under artificial and realistic noisy reverberant conditions.     

A Signal Decorrelation PNLMS Algorithm for Double-Talk Acoustic Echo Cancellation

Acoustic echo cancellation (AEC) in voiced communication systems is used to eliminate the echo which corrupts the speech signal and reduces the efficiency of signal transmission. Usually, the performance of AEC system based on the adaptive filtering degrades seriously in the presence of speech issued from the near-end speaker (double-talk). In typical AEC scenarios, double-talk detector (DTD) must be added to AEC for improving speech quality. One of the main problems in AEC with DTD is that the DTD errors can result in either large residual echo or distorting the near-end input speech. Considering the strong correlation property of speech signals, this paper presents a novel proportionate decorrelation normalized least-mean-square (PDNLMS) adaptive AEC without DTD for echo cancellation as an interesting alternative to the typical AEC with DTDs. Unlike traditional AEC with a DTD, the proposed PDNLMS uses the difference of near-end speech as the residual error to update adaptive echo channel filter during the periods of double-talk, which can efficiently reduce the double-talk influence on the AEC adaptation process. The experimental results show that not only the proposed PDNLMS without DTD illustrate better stability and faster convergence rate, but it is also of a lower steady-state misalignment and better residual signal than current methods with DTDs at a lower computational cost.     

Multichannel Online Blind Speech Dereverberation with Marginalization of Static Observation Parameters in a Rao-Blackwellized Particle Filter

Room reverberation leads to reduced intelligibility of audio signals and spectral coloration of audio signals. Enhancement of acoustic signals is thus crucial for high-quality audio and scene analysis applications. Multiple sensors can be used to exploit statistical evidence from multiple observations of the same event to improve enhancement. Whilst traditional beamforming techniques suffer from interfering reverberant reflections with the beam path, other approaches to dereverberation often require at least partial knowledge of the room impulse response which is not available in practice, or rely on inverse filtering of a channel estimate to obtain a clean speech estimate, resulting in difficulties with non-minimum phase acoustic impulse responses. This paper proposes a multi-sensor approach to blind dereverberation in which both the source signal and acoustic channel are directly estimated from the distorted observations using their optimal estimators. The remaining model parameters are sampled from hypothesis distributions using a particle filter, thus facilitating real-time dereverberation. This approach was previously successfully applied to single-sensor blind dereverberation. In this paper, the single-channel approach is extended to multiple sensors. Performance improvements due to the use of multiple sensors are demonstrated on synthetic and baseband speech examples.     

Modulated Lapped Transform Domain Acoustic Echo Canceller with Double Talk Detector

A complete acoustic echo cancellation system with double talk detection capability is presented in this paper. The proposed system includes a new acoustic echo canceller (AEC) based on the modulated lapped transform (MLT) domain adaptive structure and a robust two-stage double talk detector (DTD) to cope with MLT domain AEC. The proposed AEC achieves better signal decorrelation via orthogonal MLT of size 2N× N rather than N× N square orthogonal transform such as DCT, DFT, etc. Both the input signal and the desired response are modulated lapped transformed in order to reduce the adaptation error between them so that the signal adaptation is purely operated in MLT domain. As a complementary of this, a two-stage DTD is developed to stabilize the operation of the AEC. The proposed DTD has robust algorithm structure and it allows faster switching according to the talker state change.Several simulation results with a synthetic and real speech are presented to demonstrate the performance of the proposed AEC and DTD. The proposed MLT based AEC proven to be very useful for the echo cancellation applications requiring high convergence speed and good echo attenuation. It can achieves faster convergence rate by more than twice over those of traditional DCT based AEC with an additional advantage of 10–15 dB ERLE improvement. On the other hand, a proposed two-stage DTD is shown to react quickly to both the onset and the end of the double-talk with reasonable high accuracy.     

强海水混响背景下水中兵器攻击目标检测研究

水中兵器在对舰船目标攻击过程中受到海水混响杂波的影响,导致目标检测性能下降。传统方法采用时频分析方法进行目标检测,在信混比较低的情况下受到的干扰较强,导致虚警概率较高。提出一种舰船目标回波盲源分离的强海水混响背景下的水中兵器攻击目标检测算法。首先构建强海水混响干扰下的舰船目标回波模型,对舰船目标回波模型进行自相关匹配滤波,提取回波信号的高阶谱特征,实现目标信号的盲源分离,达到目标检测的目的。仿真结果表明,采用该算法进行目标检测,准确检测概率较高,降低了虚警概率,提高了水中兵器对目标的准确打击能力。     

语音信号去混响原理与技术

语音信号去混响技术在通信、语言识别等方面有重要应用。介绍了国内外相关研究动态和方法,阐述了声音混响过程和倒谱法去混响原理,简要介绍了传声器阵列-倒谱法去混响技术。