用于语音识别的鲁棒自适应麦克风阵列算法

对现实环境中存在的混响以及非平稳干扰语音信源等因素导致的算法性能下降,提出了一种用于语音识别的鲁棒旁瓣对消算法。讨论了旁瓣对消算法在自适应麦克风阵列中的应用,分析了算法在不同的混响条件下、不同的干扰源的噪声抑制能力。该算法通过分帧处理将输入信号划分为一系列短时平稳的信号片段。根据当前帧的信噪比决定自适应滤波器的权系数更新方式。采用一定的范数约束来限制自适应滤波器权系数的误调整。实验结果表明该麦克风阵列在混响的现实环境中能够有效抑制平稳噪声源和交叠谈话背景干扰,提高了语音识别器的抗噪性能。     

Advances in SVM-Based System Using GMM Super Vectors for Text-Independent Speaker Verification

For text-independent speaker verification, the Gaussian mixture model （GMM） using a universal background model strategy and the GMM using support vector machines are the two most commonly used methodologies. Recently, a new SVM-based speaker verification method using GMM super vectors has been proposed. This paper describes the construction of a new speaker verification system and investigates the use of nuisance attribute projection and test normalization to further enhance performance. Experiments were conducted on the core test of the 2006 NIST speaker recognition evaluation corpus. The experimental results indicate that an SVM-based speaker verification system using GMM super vectors can achieve appealing performance. With the use of nuisance attribute projection and test normalization, the system performance can be significantly improved, with improvements in the equal error rate from 7.78% to 4.92% and detection cost function from 0.0376 to 0.0251.     

Space discriminative function for microphone array robust speech recognition

0IntroductionUnder the condition of existing competing speakers,the performance of a speech recognition systemdegradesseriously.Withits capabilityto provide hands-free acqui-sition of speech and directional discrimination,micro-phone array has become widely used in many robust ASRfront-end[1-3].Adaptive beamforming realizes notches in the direc-tions of interferences in current working environment byadapting its weights according to some optimum criteri-on[4].Adaptive microphone array can re…     

基于VTS的稳健语音识别

为了进一步提高矢量Taylor级数(VTS)算法的模型补偿精度以及在噪声环境下的识别性能,提出将无监督聚类与VTS算法相结合。无监督聚类算法利用噪声模型之间的Kullback-Leibler距离将含噪语音段划分为若干个子段。然后针对各个子段分别进行一阶Taylor级数展开,并在此基础上逐段估计噪声参数和补偿声学模型。该算法结合一个中文数字串识别系统进行实验,在Babble噪声和Gauss白噪声环境下该算法的误识率相对传统的VTS算法分别下降了27.7%和17.8%。证明这种结合无监督聚类的分段VTS算法能够更加有效地将语音和噪声在倒谱域上的非线性混合模型用一阶线性模型来近似。     

Using vector Taylor series with noise clustering for speech recognition in non-stationary noisy environments

The performance of automatic speech recognizer degrades seriously when there are mismatches between the training and testing conditions. Vector Taylor Series （VTS） approach has been used to compensate mismatches caused by additive noise and convolutive channel distortion in the cepstral domain, in this paper, the conventional VTS is extended by incorporating noise clustering into its EM iteration procedure, improving its compensation effectiveness under non-stationary noisy environments. Recognition experiments under babble and exhibition noisy environments demonstrate that the new algorithm achieves 35% average error rate reduction compared with the conventional VTS.