听觉频域掩蔽效应的自适应β阶贝叶斯感知估计语音增强算法

为了在噪声抑制和语音失真中之间寻找最佳平衡,提出了一种听觉频域掩蔽效应的自适应β阶贝叶斯感知估计语音增强算法,以期提高语音增强的综合性能。算法利用了人耳的听觉掩蔽效应,根据计算得到的频域掩蔽阈自适应调整β阶贝叶斯感知估计语音增强算法中的β值,从而仅将噪声抑制在掩蔽阈之下,保留较多的语音信息,降低语音失真。并分别用客观和主观评价方式,对所提出的算法的性能进行了评估,并与原来基于信噪比的自适应β阶贝叶斯感知估计语音增强算法进行了比较。结果表明,频域掩蔽的β阶贝叶斯感知估计方法的综合客观评价结果在信噪比为-10 dB至5 dB之间时均高于基于信噪比的自适应β阶贝叶斯感知估计语音增强算法。主观评价结果也表明频域掩蔽的β阶贝叶斯感知估计方法能在尽量保留语音信息的同时,较好的抑制背景噪声。      

联合精确比值掩蔽与深度神经网络的单通道语音增强方法

针对目前有监督语音增强忽略了纯净语音、噪声与带噪语音之间的幅度谱相似性对增强效果影响等问题,提出了一种联合精确比值掩蔽(ARM)与深度神经网络(DNN)的语音增强方法。该方法利用纯净语音与带噪语音、噪声与带噪语音的幅度谱归一化互相关系数,设计了一种基于时频域理想比值掩蔽的精确比值掩蔽作为目标掩蔽;然后以纯净语音和噪声幅度谱为训练目标的DNN为基线,通过该DNN的输出来估计目标掩蔽,并对基线DNN和目标掩蔽进行联合优化,增强语音由目标掩蔽从带噪语音中估计得到;此外,考虑到纯净语音与噪声的区分性信息,采用一种区分性训练函数代替均方误差(MSE)函数作为基线DNN的目标函数,以使网络输出更加准确。实验表明,区分性训练函数提升了基线DNN以及整个联合优化网络的增强效果;在匹配噪声和不匹配噪声下,相比于其它常见DNN方法,本文方法取得了更高的平均客观语音质量评估(PESQ)和短时客观可懂度(STOI),增强后的语音保留了更多语音成分,同时对噪声的抑制效果更加明显。      

偏度最大化多通道逆滤波语音去混响研究*

房间混响会降低语音质量和语音可懂度。高阶统计量是衡量非高斯性的重要参量,基于语音非高斯特性可实现语音去混响。本文提出一种基于高阶统计量的多通道语音去混响方法,该方法首次用多通道语音信号线性预测残差的三阶统计量偏度（Skewness）构造代价函数,以去混响重建信号线性预测残差的偏度最大化为目标自适应地更新逆滤波器;同时结合语音信号的产生模型,提出基于偏度准则的线性预测与房间脉冲响应逆滤波联合估计方法,进一步提高去混响算法性能。实验结果表明,该方法相较于已有的基于线性预测残差四阶统计量峰度（Kurtosis）的方法具有更好的去混响效果,且对噪声具有更强的鲁棒性。     

基于卡尔曼滤波的低复杂度去混响算法*

在电话会议、智能音箱等应用场景下,传声器往往处在声源的远场。混响信号的存在会掩蔽后续到达的直达声信号,降低传声器接收信号的语音质量,以及语音识别系统的准确识别率。多通道线性预测算法是一种经典的盲去混响算法,但该算法往往具有较高的计算复杂度。本文提出了一种简化的卡尔曼滤波更新算法,通过对角化卡尔曼滤波器状态向量误差协方差矩阵,降低了自适应多通道线性预测去混响算法的复杂度。通过与现有分块对角简化算法对比发现,本文提出的简化算法在保证语音质量的同时,进一步降低了原卡尔曼滤波算法的复杂度。     

基于二次相关的语音信号时延估计改进算法

目前语音信号的时延估计研究,大部分采用的是广义互相关算法。然而,广义互相关时延估计算法易受噪声和混响环境影响。为此,本文提出了一种基于二次相关的语音信号时延估计改进算法,该算法对语音信号进行二次互相关运算,并结合Hilbert变换,对二次互相关峰值进行进一步的锐化处理,使得反映时延的峰值点检测更为准确。实验结果表明,改进的时延估计方法在非平稳的语音信号中能够有效地抑制噪声干扰,且在不同混响条件下时延估计具有更好的性能。     

球谐域自适应混响抵消与声源定位算法

提出了一种基于球谐域的自适应混响抵消与声源定位算法,该方法通过去混响处理改善语音质量,并提高球谐域定位算法在混响环境下的定位性能。推导了基于多通道线性预测的自适应混响抵消算法在球谐域的表达式,针对刚球模型提出分阶处理的去混响方法,并对去混响后的信号进行波达方向估计。采用32元球阵的仿真结果表明,相比于球谐域不分阶去混响方法,该方法最大可减少约2/3的运算量,同时语音PESQ得分及SRMR均显著提高。利用实验数据对算法性能进行测试,实验结果验证了该方法在实际声学环境中去混响和声源定位的有效性。      

非均匀混响统计建模与功率估计

针对非均匀混响中进行恒虚警(CFAR)检测所需的背景功率估计问题,通过对非均匀混响进行统计建模分析,提出一种非均匀混响功率自适应估计方法。所提方法基于排序统计、功率分类等处理,对混响数据服从的统计模型进行判别,根据均匀混响、参杂强干扰混响、混响边缘等具有不同特性的混响成分,选择不同的处理方式进行功率估计。进一步将所提出的混响功率估计方法应用于CFAR检测中,获得一种新的CFAR检测器。仿真和实测混响数据的处理结果表明,提出的混响功率估计方法可以对强干扰和混响边缘等非均匀混响背景进行有效判别和功率估计,提出的CFAR检测器有效提高了非均匀混响背景中的目标检测性能。      

基于噪声追踪的二值时频掩蔽到浮值掩蔽的泛化算法

虽然浮值掩蔽比二值掩蔽有更好的语音分离效果,但是由于理想浮值掩蔽难以直接估计,现有的语音分离系统通常以理想二值掩蔽估计作为计算目标。我们提出了一个二值掩蔽到浮值掩蔽的泛化算法。由于实现浮值掩蔽估计的关键在于噪声能量追踪,我们首先采用指数分布刻画以混合谱和噪声能量以混合能量及二值掩蔽为观测的条件分布。其次,采用高斯马尔柯夫条件随机场刻画噪声估计在连续几帧内的关联。最后,采用马尔柯夫链-蒙特卡洛计算噪声能量最小均方误差估计并进一步计算浮值掩蔽。实验表明,相比于基于二值掩蔽估计的常规算法,我们所提出的算法在信噪比增益和客观感知质量两方面都有显著提高。      

高分辨率有源声呐强混响抑制技术研究

有源声呐分辨率的提高可以抑制混响,但也使声呐包络数据的统计分布偏离瑞利分布,而更接近拖尾较重的K分布。强散射体所带来的强混响的幅值一般较大,它们使统计分布拖尾更严重,表征K分布的形状参数也越小。强混响作为目标干扰,严重影响了背景功率估计的准确性,从而降低了目标检测的性能。本文基于模糊统计理论,首先提出了用于抑制强混响的模糊统计归一化处理方法;然后对强混响和模糊统计归一化处理如何影响声呐数据分布和CFAR(Constant false-alarm rate)目标检测性能进行了仿真、研究和分析,最后对基于模糊统计归一化处理的CFAR检测性能和传统CFAR检测性能进行了仿真比较。仿真结果表明强混响目标干扰能使K分布数据的形状参数变小,而模糊统计归一化处理可抑制强混响目标干扰,增大包络数据分布的形状参数,提高形状参数估计性能和CFAR检测性能。      

一种基于最大似然的混响时间盲估计方法*

混响是室内声学中的重要现象,在室内设计与音频信号处理中都需要测量或估计混响时间。本文改进了一种基于最大似然估计的混响时间盲估计方法,即采用说话人在房间中自然说话时发出的混响语音信号来估计混响时间的方法。该方法首先确定语音衰减段的最优边界,其次计算该衰减段的两个额外参数,据此筛选出符合条件的语音段,最后将满足条件的语音段采用最大似然估计得到混响时间估计值。在五个不同混响时间条件下的仿真表明,与已有方法相比,改进方法估计的混响时间同真实混响时间的偏差更小,方差更低,估计准确性较高。     

Intelligibility of reverberant noisy speech with ideal binary masking

For a mixture of target speech and noise in anechoic conditions, the ideal binary mask is defined as follows: It selects the time-frequency units where target energy exceeds noise energy by a certain local threshold and cancels the other units. In this study, the definition of the ideal binary mask is extended to reverberant conditions. Given the division between early and late reflections in terms of speech intelligibility, three ideal binary masks can be defined: an ideal binary mask that uses the direct path of the target as the desired signal, an ideal binary mask that uses the direct path and early reflections of the target as the desired signal, and an ideal binary mask that uses the reverberant target as the desired signal. The effects of these ideal binary mask definitions on speech intelligibility are compared across two types of interference: speech shaped noise and concurrent female speech. As suggested by psychoacoustical studies, the ideal binary mask based on the direct path and early reflections of target speech outperforms the other masks as reverberation time increases and produces substantial reductions in terms of speech reception threshold for normal hearing listeners.     

A channel-selection criterion for suppressing reverberation in cochlear implants

Little is known about the extent to which reverberation affects speech intelligibility by cochlear implant (CI) listeners. Experiment 1 assessed CI users' performance using Institute of Electrical and Electronics Engineers (IEEE) sentences corrupted with varying degrees of reverberation. Reverberation times of 0.30, 0.60, 0.80, and 1.0 s were used. Results indicated that for all subjects tested, speech intelligibility decreased exponentially with an increase in reverberation time. A decaying-exponential model provided an excellent fit to the data. Experiment 2 evaluated (offline) a speech coding strategy for reverberation suppression using a channel-selection criterion based on the signal-to-reverberant ratio (SRR) of individual frequency channels. The SRR reflects implicitly the ratio of the energies of the signal originating from the early (and direct) reflections and the signal originating from the late reflections. Channels with SRR larger than a preset threshold were selected, while channels with SRR smaller than the threshold were zeroed out. Results in a highly reverberant scenario indicated that the proposed strategy led to substantial gains (over 60 percentage points) in speech intelligibility over the subjects' daily strategy. Further analysis indicated that the proposed channel-selection criterion reduces the temporal envelope smearing effects introduced by reverberation and also diminishes the self-masking effects responsible for flattened formants.     

Prediction of the influence of reverberation on binaural speech intelligibility in noise and in quiet

Reverberation usually degrades speech intelligibility for spatially separated speech and noise sources since spatial unmasking is reduced and late reflections decrease the fidelity of the received speech signal. The latter effect could not satisfactorily be predicted by a recently presented binaural speech intelligibility model [Beutelmann et al. (2010). J. Acoust. Soc. Am. 127, 2479-2497]. This study therefore evaluated three extensions of the model to improve its predictions: (1) an extension of the speech intelligibility index based on modulation transfer functions, (2) a correction factor based on the room acoustical quantity "definition," and (3) a separation of the speech signal into useful and detrimental parts. The predictions were compared to results of two experiments in which speech reception thresholds were measured in a reverberant room in quiet and in the presence of a noise source for listeners with normal hearing. All extensions yielded better predictions than the original model when the influence of reverberation was strong, while predictions were similar for conditions with less reverberation. Although model (3) differed substantially in the assumed interaction of binaural processing and early reflections, its predictions were very similar to model (2) that achieved the best fit to the data.     

Reverberation time and maximum background-noise level for classrooms from a comparative study of speech intelligibility metrics

Speech intelligibility metrics that take into account sound reflections in the room and the background noise have been compared, assuming diffuse sound field. Under this assumption, sound decays exponentially with a decay constant inversely proportional to reverberation time. Analytical formulas were obtained for each speech intelligibility metric providing a common basis for comparison. These formulas were applied to three sizes of rectangular classrooms. The sound source was the human voice without amplification, and background noise was taken into account by a noise-to-signal ratio. Correlations between the metrics and speech intelligibility are presented and applied to the classrooms under study. Relationships between some speech intelligibility metrics were also established. For each noise-to-signal ratio, the value of each speech intelligibility metric is maximized for a specific reverberation time. For quiet classrooms, the reverberation time that maximizes these speech intelligibility metrics is between 0.1 and 0.3 s. Speech intelligibility of 100% is possible with reverberation times up to 0.4-0.5 s and this is the recommended range. The study suggests "ideal" and "acceptable" maximum background-noise level for classrooms of 25 and 20 dB, respectively, below the voice level at 1 m in front of the talker.     

Speech dereverberation method based on spectral subtraction and spectral line enhancement

Speech signals recorded with a distant microphone usually are interfered by the spatial reverberation in the room, which severely degrades the clarity and intelligibility of speech. A speech dereverberation method based on spectral subtraction and spectral line enhancement is proposed in this paper. Following the generalized statistical reverberation model, the power spectrum of late reverberation is estimated and removed from the reverberation speech by the spectral subtraction method. Then, according to the human auditory model, a spectral line enhancement technique based on adaptive post-filtering is adopted to further eliminate the reverberant components between adjacent speech formants. The proposed method can effectively suppress the spatial reverberation and improve the auditory perception of speech. The subjective and objective evaluation results reveal that the perceptual quality of speech is greatly improved by the proposed method.     

Speech intelligibility studies in classrooms

Speech intelligibility tests and acoustical measurements were made in ten occupied classrooms. Octave-band measurements of background noise levels, early decay times, and reverberation times, as well as various early/late sound ratios, and the center time were obtained. Various octave-band useful/detrimental ratios were calculated along with the speech transmission index. The interrelationships of these measures were considered to evaluate which were most appropriate in classrooms, and the best predictors of speech intelligibility scores were identified. From these results ideal design goals for acoustical conditions for classrooms were determined either in terms of the 50-ms useful/detrimental ratios or from combinations of the reverberation time and background noise level.     

结合卷积神经网络与混响时间注意力机制的混响抑制

提出一种结合卷积神经网络的编解码器模型和混响时间注意力机制的混响抑制算法,该算法通过编解码器模型实现混响抑制,并利用混响时间注意力机制克服混响环境变化对混响抑制效果的影响。该算法在编码器中使用具有不同大小的卷积核来处理混响语音幅度谱,从而获得包含多尺度上下文信息的编码特征;通过引入注意力模块,实现在不同的混响时间环境中选择性地使用不同权重的编码特征生成加权特征;最后,在解码器中使用加权特征来重建混响抑制后的语音信号幅度谱。在模拟和真实的混响环境下,该算法相对于基线系统在语音混响调制能量比上分别取得了0.36 dB和0.66 dB的提升。实验结果表明,该算法可以适应不同混响环境的变化,相对基线系统在真实混响环境下具有更高的鲁棒性。      

Pitch-based monaural segregation of reverberant speech

In everyday listening, both background noise and reverberation degrade the speech signal. Psychoacoustic evidence suggests that human speech perception under reverberant conditions relies mostly on monaural processing. While speech segregation based on periodicity has achieved considerable progress in handling additive noise, little research in monaural segregation has been devoted to reverberant scenarios. Reverberation smears the harmonic structure of speech signals, and our evaluations using a pitch-based segregation algorithm show that an increase in the room reverberation time causes degraded performance due to weakened periodicity in the target signal. We propose a two-stage monaural separation system that combines the inverse filtering of the room impulse response corresponding to target location and a pitch-based speech segregation method. As a result of the first stage, the harmonicity of a signal arriving from target direction is partially restored while signals arriving from other directions are further smeared, and this leads to improved segregation. A systematic evaluation of the system shows that the proposed system results in considerable signal-to-noise ratio gains across different conditions. Potential applications of this system include robust automatic speech recognition and hearing aid design.     

Effects of reverberation and masking on speech intelligibility in cochlear implant simulations

Two experiments investigated the impact of reverberation and masking on speech understanding using cochlear implant (CI) simulations. Experiment 1 tested sentence recognition in quiet. Stimuli were processed with reverberation simulation (T=0.425, 0.266, 0.152, and 0.0 s) and then either processed with vocoding (6, 12, or 24 channels) or were subjected to no further processing. Reverberation alone had only a small impact on perception when as few as 12 channels of information were available. However, when the processing was limited to 6 channels, perception was extremely vulnerable to the effects of reverberation. In experiment 2, subjects listened to reverberated sentences, through 6- and 12-channel processors, in the presence of either speech-spectrum noise (SSN) or two-talker babble (TTB) at various target-to-masker ratios. The combined impact of reverberation and masking was profound, although there was no interaction between the two effects. This differs from results obtained in subjects listening to unprocessed speech where interactions between reverberation and masking have been shown to exist. A speech transmission index (STI) analysis indicated a reasonably good prediction of speech recognition performance. Unlike previous investigations, the SSN and TTB maskers produced equivalent results, raising questions about the role of informational masking in CI processed speech.     

Blind single-channel suppression of late reverberation based on perceptual reverberation modeling

A blind method for suppressing late reverberation from speech and audio signals is presented. The proposed technique operates both on the spectral and on the sub-band domains employing a single input channel. At first, a preliminary rough clean signal estimation is required and for this, any standard technique may be applied; however here the estimate is obtained through spectral subtraction. Then, an auditory masking model is employed in sub-bands to extract the reverberation masking index (RMI) which identifies signal regions with perceived alterations due to late reverberation. Utilizing a selective signal processing technique only these regions are suppressed through sub-band temporal envelope filtering based on analytical expressions. Objective and subjective measures indicate that the proposed method achieves significant late reverberation suppression for both speech and music signals over a wide range of reverberation time (RT) scenarios.