The self-organizing feature map used for speaker-independent speech recognition

Abstract

Kohonen's self-organizing feature map (SOFM) is an effective neural network for unsupervised learning. It is expected to produce a topologically correct mapping between input and output space. This paper describes a speaker-independent isolated word speech recognition system that uses a self-organizing feature map. Many experiments indicated that the self-organizing feature map algorithm shows some defects. Our speech recognition research focuses on improving the algorithm. After the improved algorithm was adopted, experimental results show that the recognition rate of this system rises significantly.