基于量子模距离的说话人识别方法

Speaker recognition method based on quantum model distance

概率测度和距离测度是模式识别最基本的两种测度,矢量量化算法是典型的基于距离测度的模式识别算法。根据量子模距离测度理论,在矢量量化算法的基础上,探索一种基于量子模距离的说话人识别方法。该方法针对说话人语音的时变性、随机性、特征维数较高等特点,将一帧语音信号视为一个量子态,并根据量子测量理论,对量子态之间进行模距离测量,从而对量子态进行有效的分类和聚类。研究表明该方法能有效地降低语音信号处理的复杂度。在经典计算机上的仿真表明,该方法在运行时间上略优于矢量量化算法,在识别率上明显优于矢量量化算法,为说话人识别的理论研究提供了新的途径。

Probability measure and distance measure are the most basic measures in pattern recogni- tion, and vector quantization is a typical pattern recognition algorithm based on distance measure. Ac- cording to the theory of quantum model distance measure, on the basis of vector quantization method, the paper explores a speaker recognition method based on quantum model distance. In order to deal with the time-varying property, the randomness and the high-dimensional characteristics of speaker＇s voice, we see a frame of speech signal as a quantum state, and measure model distance between quantum states according to quantum measurement theory. Simulation results in classical computer demonstrate that this method has slightly better in terms of operating time and visibly better in terms of recognition rate than vector quantization. It can provide a new way for the theoretical investigation on speaker recogni- tion.