Speaker Recognition based on a Novel Hybrid Algorithm

Using Probabilistic Neural Network (PNN) to recognize speaker is one of the research of branch of speaker recognition. PNN's ability of recognition is so dependent on the value of its smoothing factor that its ability of recognition is not that good. To solve this problem, we proposed a novel hybrid algorithm (DFOA-SOM-PNN) to improve PNN's ability of recognition. Firstly, it uses SOM to cluster MFCC speech characteristics parameters which can reduce storage of data and calculation, and good reflect feature of MFCC. Secondly, it uses an improved algorithm of Fruit fly Optimization Algorithm (FOA): Double group FOA (DFOA), which optimizes the smooth factor of PNN. The experimental results show that DFOA have better global convergence and fast convergence speed than FOA, and the proposed hybrid algorithm has better performance in speaker recognition.