【摘要】 应力影响下的变异语音是由于说话人受到重力加速度变化而产生的,与正常语音相比,变异语音频谱能量在频带范围内分布更加分散。把整个频带划分成8个子带,采用子带频谱能量的比值为特征,提出一种基于子空间方法的正常/变异语音分类方法。该方法采用CLAFIC方法设计初始向量子空间,并通过LSM算法对两类样本子空间按不同的旋转方式训练,用预分类的结果调整分类器的参数来改善分类器的性能。实验结果表明,该方法对应力影响下的变异语音与正常语音具有良好分类效果,平均分类正确率达到了95.9%。更多还原

【Abstract】 Speech under G-Force produces when speaker is under different accelerations of gravity,the stressful speech and normal speech have different energy distributions in the frequency domain.The whole frequency band is divided into eight sub-bands and the proportions of sub-band energy are used as classification feature.In the paper,a subspace based method is proposed to classify normal speech and speech under G-Force.The method adopts Class-Featuring Information Compression(CLAFIC) algorithm to construct initial subspaces for Learning Subspace Method(LSM),training of each subspace is rotated in different ways by LSM and classification result is used to adjust classification parameters.The experiments show that the method has good classification performance for G-Force/Normal speech and obtains the average classification rate of 95.9%.更多还原