【摘要】 利用归一化自相关函数和归一化中心二阶距两个参数,基于贝叶斯准则来识别话音和话带数据。重点讨论了加性高斯白噪声及业务先验概率估计误差对识别率的影响。实验表明:在帧长32ms,信噪比30dB时,数据的识别率为98.17%,话音的识别率达到97.29%。更多还原

【Abstract】 This paper describes an algorithm to discriminate voice and voiceband data, utilizing normalized autocorrelation and the second-order moment of the signal. The algorithm is based on Bayes Rule.It focues on discussing the influence of additive white Gaussian noise and the estimation error of service prior probability. Experiments indicate that within a window of 32ms, the ratio of correct discrimination is 98.17% for voiceband data, and 97.29% for voice.更多还原