两级决策的开集说话人辨认方法

Method of open-set speaker identification with two-level decision strategy

为了减少语音数据量 ,提高处理速度和识别的准确性 ,提出了一种采用公共码本、个人隐 Markov模型 (HMM)和个人拒识阈值进行两级决策来实现开集说话人辨认的新方法。在系统实现时 ,采用了一种改进的语音切分算法来提高输入数据的有效性 ,并将说话人识别和人脸识别融合在一起进行身份验证。实验证明这种融合方法能够有效地降低识别的相等错误率至 1%。

To reduce required speech data and improve the processing speed and the recognition precision, this paper presents a novel speaker identification method using the public codebook, the individual hidden Markov model (HMM) and the individual threshold of rejection to make a two level decision strategy. The system used an improved algorithm of speech segmentation to extract the available speech data from utterances. An approach of integrating the speaker recognition with the face recognition to verify a person's identity could further reduce the equal error rate to 1%.