【摘要】 目前,说话人识别系统对于干净语音已经达到较高的性能,但在噪声环境中,系统的性能急剧下降。一种基于子带处理的以相位自相关(PAC)系数及其能量作为特征的说话人识别方法,即宽带语音信号经Mel滤波器组后变为多个子带信号,对各个子带数据经DCT变换后提取PAC系数作为特征参数,然后对每个子带分别建立HMM模型进行识别,最后在识别概率层中将HMM得出的结果相结合之后得到最终的识别结果。实验表明,该方法在不同信噪比噪声和无噪声情况下的识别性能都有很大提高。更多还原

【Abstract】 Recently, speaker recognition system has already achieved high performance for clean speech, but in noisy environment, the performance of the system may degrade seriously. A method of speaker recognition based on sub-band processing and using phase autocorrelation (PAC) along with its energy as features is proposed. In this method, wideband speech signal is filtered into several sub-bands through Mel filter bank, features of PAC coefficient are extracted by DCT transformation, then feature vectors are modeled and identified on each sub-bands by HMM. At last, outputs from each sub-band are combined at recognition probability level. The experiment results show that the performance of recognition is improved in different SNR noisy and clean condition.更多还原