混合语音的盲分离方法研究

This project studies the blind separation problem of mixed speech using.blind source separation technique. The sufficient and necessary conditions are.presented based on 2- and higher-order cumulants for stationary and nonstationary. A.union form of several blind separation algorithms is given. New algorithms of blind.separation of instantaneous mixed signal are proposed based on 2- and higher-order.cumulants respectively. New algorithms of blind separation of convolutely mixed.signal are also proposed based on 2- and higher-order cumulants respectively. These.algorithms are used for mixed speech separation and image cryptosystem.

构造适于混合语音分离的多通道模型，提出语音盲分离的特征分解方法、离阶统计量方法、神经网络方法和信息论方法，分析算法的收敛性和多通道系统的稳定性，去除噪声和房间混响对分离效果的影响，建立语音盲分离实时处理系统。本项目将开拓语音处理的新领域，建立新理论和新方法，促进语音盲分离技术的进一步发展，具有重要理论意义和实用价值。